fix: AudiobookShelf push sends lastUpdate; 60 s per-level fetch throttle

[GianniCarlo]

Two small follow-ups from the media-server progress work (#129, #130).

Change

1. fix: the AudiobookShelf progress push sends its play date as lastUpdate — the push sent position and finished state only, so ABS stamped each entry with the moment the push landed, not when the position was reached (later than the real play for a push delayed by the queue or offline listening). AudiobookshelfProgressRequest gains lastUpdate (epoch ms), filled from the task payload's lastPlayDate — the same date the Jellyfin push sends as LastPlayedDate since fix: Jellyfin progress push reaches the server, with its play date #130. A task queued before the date was recorded omits it (Gson skips nulls) and ABS stamps its own time, as before.
2. fix: throttle each library level's contents fetch to once per 60 seconds — the per-level fetch was throttled at 30 s in two places over the same timestamp map (canFetchContents from LibraryScreen, checkAndMarkFetchContents in createFetchContentsTask). Both use one 60 s constant now, matching iOS's per-level list sync. The sort-preferences pull, documented as sharing the contents fetch's cadence, moves with it; the account-wide identifiers sync keeps its separate 30 s throttle.

Verification

AudiobookShelf 2.37 (throwaway container), emulator bp-api36, a book downloaded from that server:

Push	local lastPlayDate	ABS lastUpdate
Play start — first write, creates the entry	1790768212663	1790768213679 (ABS's clock: the first write ignores a client value)
Pause — updates the entry	1790768224474	1790768224474 (exact)

Tests. ./gradlew assembleDevDebug testDevDebugUnitTest :core:testDebugUnitTest lintDevDebug green (746 tests: 217 app, 68 wear, 461 core). New: the ABS push carries lastUpdate from the payload and omits it for a dateless task (ExternalUpdateProcessorTest); a level can't refetch within 60 s (past the old 30 s), the throttle is per level, check-and-mark and the preferences pull share the 60 s cadence (SyncStatusManagerThrottleTest, with a clock seam and a reset). Setting the throttle back to 30 s or dropping lastUpdate fails exactly those tests.

Found along the way (not in this PR)

• ABS downloads land as an unplayable zip named .mp3. ABS's api/items/{id}/download returns a zip for a book in its own folder; the download is named originalFileName ?: "<title>.mp3" (ABS list items are minified, so it's always the fallback) and archive expansion goes by that name, so it's never unzipped. iOS names downloads from Content-Disposition ("Book.zip") instead. Separate PR.
• Every ABS server gets the stable id server-settings, the constant ABS reports as serverSettings.id, so two ABS servers can't be told apart by host id.

[github-actions]

✅ Claude PR Review — PASS

This PR does two things.

• AudiobookShelf push: the progress push now sends lastUpdate in epoch ms, taken from the task payload's lastPlayDate. That is the same millisecond value the Jellyfin push already turns into LastPlayedDate. The field is nullable and defaults to null, and the :core Gson setup doesn't enable serializeNulls, so tasks queued before the date was recorded still leave the field out.
• Fetch throttle: the per-level contents fetch and the preferences pull now share one 60 s FETCH_THROTTLE_MS constant. A clock hook was added for tests. The account-wide identifiers throttle stays at 30 s on purpose.

I checked the callers and the payload producer (SyncTaskFactory stores lastPlayDate as epoch ms). Both new test files cover the behaviour and put SyncStatusManager's global state back afterwards. Open finding #1 looks fixed (@VisibleForTesting and @Volatile are now applied). I found no new problems.

Findings: no findings

Previously raised

Finding	Status
core/src/main/java/com/tortugapower/audiobookplayer/logic/SyncStatusManager.kt:33 (info)	✅ verified fixed in a93c89c

Converged: nothing new this round, and every earlier finding is settled.

_{Model claude-opus-5-5 · run log · 0 new · 0 carried over · 1 verified closed · 0 resolved · advisory (a human should still review). Findings are de-duplicated across pushes; an earlier finding closes only when the verification pass judges it against the current code — fixed, no longer applicable, accepted by a maintainer, or a duplicate of a finding reported on this push.}