refactor(github): generalize property fetching, fix cancellation swallowing

[PonceGL]

What

GitHubAnnouncementPropertiesService.fetchPlayStoreAnnouncement() mixed two
concerns: the generic mechanics of reading a raw .properties file from a
GitHub repo, and the Play Store announcement's own field mapping. Splitting
them lets other consumers (a remote feature-flag file, for instance) read
their own config without knowing anything about announcements. One of the
small independent fixes listed in #2813.

While generalizing it, found and fixed a real bug: the catch wrapped
Exception broadly, silently turning a coroutine cancellation into an
ordinary Result.failure instead of letting it propagate.

Change

• fetchProperties(owner, repo, branch, configPath): public, generic, no
  defaults (a generic utility shouldn't default to any one caller's repo).
• Split into three internal pieces: buildRawContentUrl (pure URL
  construction), fetchRawProperties(rawUrl) (the actual HTTP call),
  Properties.toPlayStoreAnnouncementRemoteConfig() (the mapping).
  fetchProperties() itself always points at real GitHub with no way to
  intercept it in a test — fetchRawProperties accepts any URL, so tests
  point it at a local com.sun.net.httpserver.HttpServer (JDK, no new
  dependency) instead.
• fetchPlayStoreAnnouncement() keeps its exact behavior and defaults
  (including the pre-rename repo name — that's a separate, already-open PR,
  not this one).
• catch (e: CancellationException) { throw e } before the broad catch —
  the actual fix.

Honest about what the cancellation fix does and doesn't guarantee:
HttpURLConnection is a classic blocking-socket API and does not react to
Thread.interrupt() — verified empirically (tried runInterruptible and a
disconnect() triggered from invokeOnCompletion; neither achieves real
cancellation of a call already blocked in getResponseCode()). What this
fix guarantees is narrower but real: cancellation is never swallowed once
it does reach this code. Making the blocking call itself promptly
interruptible would mean replacing HttpURLConnection with something that
supports it (OkHttp) — out of scope here, noted in the code for whoever
touches this next.

Testing

10 JVM tests, stable across repeated runs. The cancellation test is
deterministic (cancel the coroutine before the call starts) rather than
timing-dependent, for the reason above — a during-flight cancellation test
would be flaky by construction against a blocking API that doesn't actually
interrupt.

assembleDebug succeeds, full JVM baseline unaffected (5 pre-existing
failures, none new).

[PonceGL]

Closing for now — reorganizing how this work is staged. It'll go through our fork first and we'll propose it upstream again, possibly bundled differently, once the larger feature it's part of is further along. Not a rejection, just a process change on our side.