fix(sandbox): bound quick file ops during link partitions

[tlgimenes]

Summary

• Add a 10s fast-fail timeout for quick sandbox file operations when the daemon link is partitioned.
• Update the WS partition resilience scenario timing expectations.
• Align the Playwright config unit test with MCP_CACHE_ENABLED=true.

Test Plan

• bunx biome format --write apps/mesh/src/api/routes/sandbox-proxy.ts tests/resilience/scenarios/link-dispatch-ws-partition.test.ts apps/mesh/playwright.config.test.ts
• bun run check
• bun test apps/mesh/playwright.config.test.ts

Note: full bun test was not completed in the sandbox because Docker-backed resilience setup did not progress.

---

Summary by cubic

Bound quick sandbox file ops to a 10s fast-fail so reads/writes don’t stall ~30s during link partitions and instead return a prompt 502. Updated resilience test timing and aligned Playwright config test with MCP_CACHE_ENABLED=true.

• Bug Fixes
  • Added a 10s QUICK_FILE_OP_TIMEOUT_MS and quickFileOpSignal; applied to read/write/mkdir/unlink/rename/glob routes to fail fast on partitions.
  • Left streaming ops (exec, events, git/*) unbounded to preserve behavior.
  • Resilience test: assert quick-file-op errors within ~10s (15s ceiling) and extend reconnect waits to 90s for CI backoff.
  • Playwright config test: expect MCP_CACHE_ENABLED=true in the dev server command.

^{Written for commit 3d7c9cf. Summary will update on new commits.}