Let the Linux head check for updates, and seal its device key in the TPM

Two things the RPM shipped without, and the reason they arrive together is
that both are the same question — what may this platform ask of the machine,
and what must it not ask of the user.

◆ UPDATES: THIS CLIENT CHECKS AND FETCHES, AND PACKAGEKIT INSTALLS.

ADR 0013 rule 11 said there was no updater at all, and that went too far. It
conflated noticing that a newer build exists with installing it: the second is
what the trust argument is about, and the first is not — a client that checks
a fixed address and tells you what it found has taken no capability from
anybody. Written as a flat refusal, the rule bought its safety by paying a cost
that was avoidable, and the ADR now says so in its own words.

So LinuxUpdateChannel reads the rolling nightly release, compares versions the
way rpm compares them, downloads the package, verifies its digest and hands the
file to PackageKit over gdbus. Polkit asks; the user may say no; nothing is
torn down before the answer, because ApplyingEndsTheProcess is false and the
shell already knows what that means — it is the phone's shape, and declining an
update must not cost somebody their open shells.

Three details that are load-bearing rather than incidental:

  - The comparison is rpmvercmp and not SemVer, because rpm is what will accept
    or refuse the install. Same argument ADR 0014 makes for versionCode. The
    installed side is read out of the rpm database rather than derived from the
    assembly, so the SemVer-to-rpm translation exists in exactly one place. All
    28 expected values in RpmVersionTests came out of rpm itself rather than out
    of a reading of the algorithm.

  - The digest is checked and is not a signature. Velopack verifies its own
    downloads and Android checks a signature; an RPM over HTTPS has neither. The
    manifest names a SHA-256 and a mismatch deletes the file rather than leaving
    something installable-by-hand in the profile directory.

  - ONLY_TRUSTED is cleared, which is what routes the request to polkit's
    package-install-untrusted action. The package is unsigned, so requiring
    trust would fail every update at the last step.

◆ THE DEVICE KEY: SEALED IN THE TPM, BEHIND A PIN THE TPM CHECKS.

ADR 0007's amendment is the bar: a gate inside the process is not a gate. The
key is sealed into a TPM 2.0 keyedhash object whose userAuth comes from a PIN,
so unsealing is a command the chip refuses; the blob is encrypted to the storage
hierarchy and is useless on any other machine; nothing here ever compares a PIN.

The sealed object deliberately carries no NoDA attribute, and that one omission
is the whole reason four digits is enough: every wrong answer advances the TPM's
dictionary-attack counter and the chip stops answering entirely once it trips.
Measured at four wrong tries against a software TPM, after which the correct PIN
gets nothing either. It is asserted by a test rather than left to a reading of
the template, because removing the attribute leaves every other assertion
passing.

This is weaker than Hello and Touch ID and the difference is invisible, so it is
written down in three places rather than smoothed over: those raise their prompt
from inside the platform and this one is a window we draw. The prompt is
deliberately plain for that reason — teaching people that a convincing DodoSSH
PIN box is a system dialog is the one thing this design must not do.

Rejected, each for failing the amendment: the Secret Service API, which has no
per-use consent and is DPAPI's shape exactly; fprintd, which answers a question
and holds nothing, so skipping the call skips the gate; and a polkit-gated root
helper, which would work and puts a privileged component in a package that
installs nothing setuid.

Availability is narrower than "has a TPM": on Fedora /dev/tpmrm0 is root:tss and
desktop users are not in tss, so the ordinary case reports unavailable and keeps
asking for the passphrase. Shipping a udev rule to widen that is refused — it is
the administrator's decision, not an application package's.

◆ WHAT IS ACTUALLY VERIFIED, SINCE THAT IS THE PART THAT USUALLY ISN'T.

This is the only device key store in the repository with real coverage, and it
is only possible because the PIN is a delegate and the TPM is a device: seal,
unseal, wrong PIN, lockout, decline and a foreign blob all run unattended
against a software TPM. Windows and macOS cannot be tested at all — their
prompts block a suite forever, which WindowsDeviceKeyStoreTests records.

TSS.Net's own TcpTpmDevice cannot reach swtpm: it opens the platform socket too
and exchanges a handshake only the Microsoft reference simulator answers, so
Connect() blocks with no error and no timeout. SoftwareTpm speaks the data
channel and nothing else.

The suite also lied once and was fixed: the lockout test trips shared state, and
without a DictionaryAttackLockReset in Dispose every later test got null from a
correct PIN — six of them assert null and passed for entirely the wrong reason.

End to end on a real Fedora 44: a genuine linux-x64 publish packages, installs,
and every DT_NEEDED in the installed tree resolves except liblttng-ust, which is
the one this spec deliberately excludes. What no runner covers is /dev/tpmrm0,
PackageKit and polkit, and Phase 19 grew five checks for exactly those.

Also fixed: the magic-bytes literal picked up raw control characters, which made
LinuxDeviceKeyStore.cs binary to git — no diffs and no line-ending normalisation.

docs/adr/0013-desktop-distribution-and-updates.md

@@ -323,20 +323,54 @@ named the two platforms it serves rather than negating the one it does not, on t
 AppImage should be an addition here rather than a thing that silently already half-happened". This is that
 addition, and it is not an AppImage.
 
-**The package is an RPM and there is no updater in it.** That is the decision, and everything else follows
-from it. Velopack does have a Linux path, so the shape of decisions 1 through 9 could have been repeated a
-third time — one more channel, one more index, one more client polling the forge. It should not be, because
-this platform already has the mechanism that whole apparatus is a substitute for. `dnf` is an updater the
-user already trusts, already audits and already controls, and it updates the runtime's own dependencies
-besides. Adding a second one that answers to this project instead would be building the capability decision
-4 spends a page keeping unsettable, on the one platform where nobody had to.
+**The package is an RPM and nothing inside it replaces anything.** That is the decision, and everything
+else follows from it. Velopack does have a Linux path, so the shape of decisions 1 through 9 could have been
+repeated a third time — one more channel, one more index, one more client swapping its own files over. It
+should not be, because this platform already has the mechanism that whole apparatus is a substitute for.
+`dnf` is an updater the user already trusts, already audits and already controls, and it updates the
+runtime's own dependencies besides. Adding a second one that answers to this project instead would be
+building the capability decision 4 spends a page keeping unsettable, on the one platform where nobody had
+to.
 
-The honest cost, recorded because it is the part that will be argued with: **discovery is worse here than on
-either other desktop.** Windows and macOS clients learn about a new build within six hours without anybody
-deciding to look. A Linux user learns when their repository does, and this project publishes no repository,
-so today they learn when they next go and look. That is a real regression against the sharpest thing this
-ADR bought, accepted because the alternative regression — a second update path with a different authority
-behind it — is the one the trust model cannot afford. A package repository is the fix and is not built.
+◆ **The first version of this rule said "there is no updater in it at all", and that went too far.** It
+conflated two things the other two platforms happen to do together: *noticing* that a newer build exists,
+and *installing* it. The second is what the argument above is about, and the first is not — a client that
+checks a fixed address and tells you what it found has taken no capability from anybody. Written as a flat
+refusal, the rule bought its safety by paying the cost in the paragraph below, and the cost was avoidable.
+
+**So the split is: this client checks and fetches, and PackageKit installs.** `LinuxUpdateChannel` reads
+the same rolling release the Windows nightly feed lives on, compares versions the way `rpm` compares them,
+downloads the package, verifies its digest, and hands the file to PackageKit — which raises polkit and asks
+the user, who may say no. `ApplyingEndsTheProcess` is false and nothing is torn down before the answer, which
+is the phone's shape rather than the desktop's, and is already what `MainWindowViewModel` does with that
+flag. Four things follow that are worth stating because each is a property the Velopack channels do not have
+to think about:
+
+- **The comparison is `rpmvercmp` and not SemVer.** Decision 7 made one version of record for the whole
+  repository, and this is the one place it is not the string that decides: `rpm` will accept or refuse the
+  install, so the client must compare the way `rpm` does or it can offer an update PackageKit then declines.
+  It is the same argument [ADR 0014](0014-android-updates.md) makes for comparing `versionCode`. The
+  manifest therefore carries both — the RPM version to compare and the SemVer to show a person — and the
+  installed side is read out of the rpm database rather than derived from the assembly, so the translation
+  `scripts/ci-linux-rpm.sh` performs exists in exactly one place.
+- **The digest is checked, and it is not a signature.** Velopack verifies what it downloads and Android
+  refuses a package whose signature does not match the installed one. An RPM fetched over HTTPS and handed
+  to a package manager has neither, so the manifest names a SHA-256 and the client refuses anything else.
+  That catches a truncated download and a caching proxy and says nothing whatever about provenance, because
+  the same job writes the package and the manifest. Provenance here is decision 4's constant address.
+- **`ONLY_TRUSTED` is cleared, deliberately.** The package is unsigned, so requiring PackageKit to trust it
+  would fail every update at the last step. Clearing the flag routes the request to polkit's
+  `package-install-untrusted` action, which asks for an administrator rather than proceeding quietly — the
+  user is told what they are agreeing to, which is the honest form of an unsigned package.
+- **Nothing restarts.** Replacing files under a running process is ordinary on Linux and the old inodes stay
+  mapped, so the build keeps running until somebody closes it. That is the same courtesy decision 1's
+  consequences argue for on Windows, arrived at for free.
+
+The honest cost that remains, recorded because it is the part that will be argued with: **discovery is still
+worse here than on either other desktop, for anybody who declines the in-app check.** There is no package
+repository to subscribe to, so `dnf upgrade` on its own will never bring a newer DodoSSH; the in-app check is
+what stands in for one, and a user who turns it off on PREFERENCES → UPDATES has no other signal. A signed
+repository is the fix, it is not built, and decision 11's last paragraph says what it would cost.
 
 **CI publishes it, and only to the nightly channel.** Rule 3 is untouched and applies with its full force:
 the token that writes a release is the capability to ship somebody a build, and it stays off runners. What
@@ -360,10 +394,15 @@ different paths cannot replace one another and `dnf` says so, where on Windows t
 built out of pack ids. The profile split is unchanged and still belongs to the build: `DodoChannel=nightly`
 is what `ClientPaths` reads, and the packaging must not answer that question a second time.
 
-**Two things this platform does not have, and they are stated rather than left to be discovered.** There is
-no hardware-backed device key store — `DesktopDeviceKeyStores` returns `UnavailableDeviceKeyStore`, so the
-passphrase is asked for at every unlock, where Windows has a TPM and a Mac has the Secure Enclave. And there
-is no signature on the package, which is not decision 8's "unsigned for now": nothing on this platform
+**The device key store is settled separately and is not nothing.** The first version of this rule said this
+platform had none and that the passphrase was asked for at every unlock. It has one:
+[ADR 0007](0007-device-key-protection.md) decision 4 adds a TPM-sealed key released against a PIN the TPM
+itself checks and counts failures against. It is weaker than Hello and Touch ID in a way that ADR names
+precisely, and it is available on fewer machines than have a TPM, because the ordinary Fedora desktop user
+may not open `/dev/tpmrm0`.
+
+**What this platform does not have is a signature on the package**, which is not decision 8's "unsigned for
+now": nothing on this platform
 refuses or warns about an unsigned RPM, so there is no dialog to buy a certificate to remove. What stands in
 for one is the same thing that stands in for it on the forge — TLS, and a project that publishes from one
 address. An RPM repository would change that, because repository metadata is signed and a repository is a