aw-sync: rowid cursor misses heartbeat extensions to trailing open event #790

Problem

The rowid-cursor sync introduced in #789 doesn't re-fetch events whose endtime was extended by a heartbeat after the event's rowid was synced.

Scenario:

1. Event A (rowid 100, T1–T2) is synced → cursor saved at 100
2. A heartbeat extends event A's endtime to T2+30 (rowid still 100)
3. Next sync: get_events_since_rowid(100) returns only rowid > 100 → misses the T2→T2+30 extension
4. reconcile_updated_events handles the lookback window up to resume_sync_at but not extensions beyond that boundary

Impact

Bounded: only the trailing open event at sync time loses the heartbeat extension in the destination. The impact resolves on the next natural event transition (when a new event starts and the old one closes).

Tradeoff

This is a known tradeoff vs. the old timestamp-scan path:

• Rowid cursor: correct for late-arriving events (the core aw-sync: no stored cursor — resume-from-destination silently never syncs late-arriving events #696 fix), but misses endtime-only updates
• Timestamp scan: re-fetches events in the time window, catching heartbeat extensions, but permanently loses late-arriving events (aw-sync: no stored cursor — resume-from-destination silently never syncs late-arriving events #696)

Proposed fix

After the rowid fast-path, re-sync the highest-rowid event from source (the "trailing event") to capture any heartbeat extension since the previous cursor save. This is a small fixed-cost read per bucket per sync pass.

Related

• Fixes aw-sync: no stored cursor — resume-from-destination silently never syncs late-arriving events #696 (rowid cursor) — fix(aw-sync): replace destination-as-marker cursor with per-source rowid cursor #789
• Greptile P1 finding "Continued activity stops growing" on fix(aw-sync): replace destination-as-marker cursor with per-source rowid cursor #789