fix: stop reporting "Sync 100%" while a wallet rescan is owed (#103)

[jvsena42]

Summary

Fixes P2 of #103, plus a worse sibling found while auditing it: the foreground-service notification and the Node screen disagreed about how synced we are.

P2. SettingsViewModel arms WALLET_NEEDS_RESCAN on descriptor load (:451) and birthday change (:313), but FlorestaService only fires the rescan 60s after filters reach the tip. The stepper's allDone consulted only rescan_in_progress, so for at least that minute — plus up to 10s of poll latency — the UI reported 100% with a rescan still pending. That is what put @TheButterZone in a position to reconnect their Electrum wallets at step 20 when it wasn't yet safe to.

The notification was worse. applySyncState knew only headers → blocks → stalled → synced. With no filters phase and no rescan phase at all, it read Synced - Block #N / "Fully synced" through the entire compact-filter sync — ~3 hours in the reporter's log (steps 11→13) — and through a running rescan, while the Node screen correctly said "Syncing filters" / "Scanning wallet".

The root cause of the drift: the phase logic existed in three hand-rolled copies (ScreenNode, ScreenNodeTablet, FlorestaService).

What changed

• domain/floresta/SyncPhase.kt — one pure phase computation shared by all three surfaces. WALLET_SCAN now covers a rescan that is pending as well as one in flight. Precedence preserves today's phone-screen title order exactly. Also holds computeFilterSyncDecimal / computeRescanProgressDecimal, so the screen and the service stop computing them differently.
• presentation/service/SyncNotificationContent.kt — maps a phase to notification text, free of Android types so it is unit-testable. synced = (phase == SYNCED) is the one line that makes "Fully synced" impossible during filters or a rescan.
• domain/floresta/WalletRescanGate.kt — the service's grace-window state, plus a retry bound. Since the UI now stays unsynced while the flag is set, a rescan that can never succeed (e.g. NoBlockFilters) would otherwise pin it there forever. Ten attempts, backoff doubling from 60s to a 30m cap, then it clears the flag. Timing uses a monotonic TimeSource, not the wall clock — an NTP correction mid-session would otherwise make a long backoff fire instantly or never.
• ScreenNodeTablet now takes a SyncPhase instead of five derived booleans, so it can't be handed an inconsistent set.

Note rescanblockchain flips rescan_in_progress synchronously before returning Ok, so clearing the flag right after a successful RPC leaves no gap where both pending and in-progress read false.

Tests

42 new tests (331 total). Beyond the per-phase cases, SyncNotificationParityTest walks all 256 combinations of ibd × progress × filters × rescan × pending × stalled and asserts the notification's synced flag equals the stepper's allDone — that is what fails today and what stops the two surfaces drifting again.

Reverting just the P2 condition (rescanInProgress || rescanPending → rescanInProgress) fails exactly four tests, one per surface, confirming they are load-bearing.

Verified on device

Mainnet emulator at height 959,254. With ibd=false, progress=1.0 and filters at 2.5%:

Surface	Before	After
Node screen	Syncing filters 2.83%	Syncing filters 2.83%
Notification	Synced - Block #959,254 / Fully synced	Syncing filters: 2.52%

Also confirmed WALLET_NEEDS_RESCAN arms on descriptor load (read back from DataStore), and that with it armed while filters are at 12.67% both surfaces still correctly show "Syncing filters" — a pending rescan must not preempt a phase that has to finish first.

Scope

P1, P3 and P4 of #103 remain open; P1 and P4 are filed upstream (getfloresta/Floresta#1225, #1226).

Checklist

• Commits are atomic and follow conventional style (feat:, fix:, chore:, ...)
• Unit tests added/updated for business-logic changes
• Journeys and journeys/README.md testTags updated for UI changes — N/A, no testTags added or renamed
• ./gradlew detekt, ./gradlew lintDebug, and ./gradlew test pass locally
• I have self-reviewed my own diff

Preview

Text-only change to sync labels; the before/after notification text is in the table above.