improve observability and readiness timeout for buildkitd startup (v2) - #113

Summary

Ports #112 (merged to main / v1) onto release/v2. When buildkitd forks successfully but its gRPC workers aren't ready in time, the action falls back to a cold local builder without leaving any evidence of what the daemon was doing. This changes the worker-readiness timeout path to capture evidence and report the failure, and gives slow inits more time before falling back:

• Raise buildkitdTimeoutMs from 30s → 120s: worker init can be slow when loading a large cache from the sticky disk, and a cold local-builder fallback costs far more than waiting.
• On worker-readiness failure, tail the last 100 lines of /tmp/buildkitd.log into the job log (new logBuildkitdLogTail(), also reused by the post-step crash-log path) and call reporter.reportBuildPushActionFailure("BUILDER_STARTUP", ...) so the backend sees these fallbacks.
• Report BPA_BUILDKITD_READY_DURATION_MS in a finally block so the metric is emitted on failure too, not just success.

Cherry-pick of 5fe3b77 with dist/ rebuilt against v2.

Link to Devin session: https://app.devin.ai/sessions/f9e26ad52787476784fe0848cd40551d
Requested by: @ajwerner

^{Need help on this PR? Tag /codesmith with what you need. Autofix is disabled.}

^{Need help on this PR? Tag /codesmith with what you need. Autofix is disabled. (Staging)}