fix(oura): publish the ring's history drain to LiveState.backfilling so the sync indicators light under a ring

[pipiche38]

Every sync indicator — the Today header capsule and status light, the sync chip, the Sleep, Live and Health "Syncing…" states, the macOS menu-bar row and the #1164 "Pending sync" caption on today's Rest — reads ONE flag, LiveState.backfilling, with syncChunksThisSession beside it. Only BLEManager (the WHOOP offload) ever raised it. The Oura drain stamped lastSyncedAt at completion and nothing else, so under a ring all of those surfaces stayed at rest through every drain: a wearer whose ring was handing over a night saw the same header as one whose ring was idle, and the "handing over history now" beat #245 added to the header never fired for them.

Change

• Swift OuraLiveSource: enterBackfilling() at fetch start (chunk tally reset, mirroring BLEManager.startBackfilling), one chunk per 0x11 batch summary, exitBackfilling() in finishDrain on every end (caught up, stalled, deadline, no cursor progress) before the completed-offload stamp, so the re-score that stamp triggers sees the flag at rest and does not defer Today's history-wide reads (Community and team-chat notification parity: likely out of scope? #755). A publishedBackfilling latch means the ring only ever lowers a flag it raised — LiveState is one object every source writes into.
• The five duplicated live.connected = false; live.streamingLiveHR = false link-down writes fold into one markLinkDown() that also closes a drain cut by the drop, so a flag cannot be left up holding the header in "Syncing" until the next drain.
• Kotlin twin: WhoopBleClient.publishExternalBackfilling(active, chunks) beside publishExternalBattery, a syncSink on SourceCoordinator wired in NoopApplication, and an onBackfilling callback on OuraLiveSource raised / ticked / lowered at the same four points (fetch start, batch summary, finishDrain, stop() + STATE_DISCONNECTED).

Behaviour to expect

A caught-up ring's periodic drain is one GetEvents round trip at bytes_left 0, so the capsule expands and settles within seconds; a drain with a night behind it holds it up for the ~1–2 min a full pull takes, chunk tally ticking. The chained pass after a deadline stop lowers and re-raises it 5 s apart; ChargeSyncIndicator already handles a restart mid wind-down. Every action gated on the flag (onAbortSync, Health's "Sync now") is additionally WHOOP-gated, so nothing strap-only is exposed under a ring.

Not in this PR, noted for parity: the Android Oura drain still stamps no lastSyncAt at completion (the Swift side has since 2026-08), so Android's sync chip reads ✓ from the WHOOP timestamp only. Separate concern.

Verification

• Strand (macOS) builds; compileFullDebugKotlin clean; doc_comment_lint green.
• The drain is BLE-driven end to end, so the flag's timing is only checkable on hardware: Hardware: owed — to be confirmed on a ring against the strap log's fetching history from cursor / history caught up bracket with the Today header watched during it.

Refs #245, #2208.