fix: shorten write transaction lifetimes to resolve sync lock contention

src/runtime/watch-pipeline-ingest.ts

@@ -133,6 +133,7 @@ export function createWatchIngest(options: WatchIngestOptions): WatchIngest {
     poll_interval_ms: cfg.watch.poll_interval_ms,
     on_checkpoint: (ts) => {
       db.setState('file_watcher.checkpoint_ts', String(ts))
+      db.flush()
     },
   })
 

src/runtime/watch-pipeline-processor.ts

@@ -67,6 +67,7 @@ export function createWatchPipelineProcessor(ingest: WatchIngest): WatchPipeline
         try {
           await db.insert(enriched)
           db.setState('events.last_ts', String(enriched.timestamp))
+          db.flush()
           stats.stored++
         } catch (err) {
           stats.errors++

src/sync/sync-engine.ts

@@ -48,6 +48,11 @@ export class SyncEngine {
     }
 
     try {
+      // Commit any pending pipeline transaction so sync operates on a clean
+      // connection state.  This is an intentional cross-subsystem commit:
+      // each pipeline event is an independent unit, so force-committing
+      // partial pipeline work is safe and prevents write-lock contention.
+      this.options.db.flush()
       const cfg = this.options.config.sync
       if (!cfg?.server_url) {
         result.errors.push('sync.server_url not configured')