Prioritize foreground reads during background sync

[marcus-pousette]

What changed

• Adds RPC priority metadata plus foreground/background scheduling to wa-sqlite worker clients and the shared worker.
• Marks inbound sync applyOps appends as background work.
• Routes read/query APIs as foreground work so they can run between background sync apply batches.
• Keeps local writes in the normal lane so they do not overtake an active background sync stream.
• Updates the runtime benchmark to exercise background remote sync applies.
• Adds shared OPFS e2e coverage for applying remote batches around an interleaved local write.

This is stacked on #166. It builds on the batch-level yield there: #166 creates a scheduling gap between inbound sync batches; this PR decides what should run in that gap.

Performance Results

Measured on shared-worker OPFS with real SyncPeer.pushOps ingest, 2,000 remote ops, 500-op remote batches, 20 local writes, and 20 paginated reads using childrenPage(root, null, 50).

Prefill	Branch	Inter-batch read	Read p50	Read p95	Local write p50	Total duration
0	#166 baseline	36.70ms	37.00ms	n/a	38.45ms	1422ms
0	this PR	6.70ms	6.83ms	151.31ms	40.37ms	1561ms
5k	#166 baseline	160.16ms	3.07ms	n/a	347.27ms	7488ms
5k	this PR	5.61ms	5.22ms	400.82ms	369.54ms	8503ms
10k	#166 baseline	602.83ms	40.27ms	1067.58ms	570.89ms	13575ms
10k	this PR	5.96ms	4.90ms	563.58ms	562.56ms	12906ms

The main improvement is the specific read scheduled immediately after a remote batch: 160ms -> 5.6ms at 5k prefill and 603ms -> 6ms at 10k prefill. This does not make a running appendMany batch interruptible, so read p95 can still be high when a read lands behind an active batch or normal write.