Optimize wa-sqlite binary RPC reads

[marcus-pousette]

Summary

• return tree.parent and tree.getPayload worker RPC results as Uint8Array instead of number arrays
• transfer those binary response buffers for dedicated-worker and shared-worker reads
• chunk large browser appendMany calls at RPC boundaries so a single large sync/import batch does not monopolize the worker queue
• extend the existing wa-sqlite responsiveness e2e to exercise one 5k-op append while sampling child and payload reads

Why

The stall shape we care about is a read getting stuck behind a large write batch in the worker. The test now runs one large append by default, then reports child-read and payload-read p50/p95/max during that write.

Local single-run comparison against origin/main with one 5k-op append:

Runtime	main child read max	PR child read max	PR payload read max
dedicated-worker	720ms	236ms	236ms
shared-worker	802ms	537ms	493ms

The PR still keeps write throughput in the same rough range; the goal is to avoid second-scale read stalls, not to add a fragile microbenchmark threshold.

Testing

• node packages/treecrdt-wa-sqlite/scripts/copy-opfs-vfs.mjs --sources-only && ./node_modules/.bin/tsc -p packages/treecrdt-wa-sqlite/tsconfig.json
• ./node_modules/.bin/prettier --check packages/treecrdt-wa-sqlite/src/client.ts packages/treecrdt-wa-sqlite/e2e/tests/responsiveness.spec.ts packages/treecrdt-wa-sqlite/src/rpc.ts packages/treecrdt-wa-sqlite/src/common-worker.ts packages/treecrdt-wa-sqlite/src/worker.ts packages/treecrdt-wa-sqlite/src/shared-worker.ts packages/treecrdt-wa-sqlite/e2e/src/responsiveness.ts
• git diff --check
• ./node_modules/.bin/playwright test tests/responsiveness.spec.ts --project chromium-dev --reporter=line from packages/treecrdt-wa-sqlite/e2e using a local temporary pnpm shim because this shell has no pnpm binary