Handle detached provider identity on watchdog stop

[brsbl]

Summary

• treat thread.stop as complete when the runtime has already detached provider identity for the same thread
• keep normal stop failures surfaced for any other error
• add host-daemon regression coverage for detached provider identity during stop

Investigation

The 902s watchdog event was real provider inactivity: the manager accepted an automatic system tell after child thread thr_nzet85hdw8 hit a Claude Code rate limit, then emitted no provider activity for 902423ms. The recovery path then got stuck because repeated thread.stop commands failed with No provider associated with thread "thr_8wvfkbawbi".

Tests

• pnpm exec turbo run test --filter=@bb/host-daemon -- --run test/command/thread-dispatch.test.ts
• pnpm exec turbo run typecheck --filter=@bb/host-daemon