fix(gen2): harden workspace lifecycle recovery

[yousef20920]

Summary

• Recover stale Gen 2 startup claims and let concurrent opens poll the shared workspace state without holding a long server request.
• Bound each host-wake attempt and retry startup in the workspace page while Azure finishes waking.
• Fence stop/start transitions, preserve snapshots when the host is unreachable, and wake the host before deleting guest and snapshot data so teardown fits the 300-second function limit.
• Detect Firecracker exits during snapshot restore and retry from saved workspace disks; prevent new sandboxes from racing host shutdown.

Verification

• pnpm typecheck passed.
• pnpm lint passed with three existing warnings and no errors.
• pnpm test passed: 13 infrastructure checks, 1,122 web tests (one skipped), and all other workspace package tests.
• Focused lifecycle Vitest suite passed: 30 tests across four files.
• Apple Container Rust checks passed: cargo fmt --check, Clippy with warnings denied, and 79 tests.
• Local production build passed; http://127.0.0.1:3002/api/health returned HTTP 200.
• Local Playwright suite passed: 38 tests, one skipped.
• All GitHub checks passed on the current head.

Authenticated localhost lifecycle smoke

• Ran two full Playwright passes against a local production next start on port 3003, with the current Drizzle schema in a disposable Apple Container PostgreSQL instance on port 55432 and CODEV_FAKE_GUEST=1. Each pass completed two create/open/delete cycles and two stop/reopen cycles; ready state and deletion were verified after every operation.
• A concurrent lifecycle pass issued four opens together, stopped the instance, issued four reopens together, and deleted the workspace. All open requests returned 200 or 202; the final state was ready; stop and delete returned 200.
• These runs exercised the real local Next.js routes and database with the guest boundary replaced by the in-memory contract fake. They did not create an Azure VM or Firecracker guest.

Environment limits

• The preconfigured shared database in apps/web/.env.local has not applied migration 0054, which is already on main and adds deleting to gen2_workspace_status. Deletion against that stale schema returned 502 while reserving the delete state. The synthetic test account and fake workspace were removed; the shared database was not migrated. The isolated localhost lifecycle runs used the current schema.
• A live Azure-backed host wake remains unverified because the local Azure CLI refresh token has expired. The Firecracker early-exit and host-shutdown reservation regressions are covered by the Rust tests and CI.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.