fix(apple): expose toolchain probe failures in smoke builds

Merged
TH
2d ago
+144-8

Summary

Recent iOS smoke failures affect main: run 40339024659 at 840f9c40e3 compiled successfully, then exhausted the toolchain fingerprint budget before smoke scenarios ran.

Record each probe's command, budget, duration, exit, normalized exception, and classified result failure. Enable diagnostics in the metadata writer and preserve its final typed error details. Three files changed; no scope expansion. Issue: none (directly requested investigation).

This fixes missing evidence, not the intermittent hosted stall. Timeout and retry policies remain unchanged.

Validation

  • Head: 2d667a9c2fe573a555c9418ec5d5e4db7fd723aa.
  • pnpm check:affected --run --head HEAD: passed; 37 checks passed/25 skipped; 2,806 related tests passed.
  • Receipt run-9nLp5I: status=passed, exactCommit=true, base 581357bc72c65de9ba798e7299681b19a5c13af4.
  • Focused regression command on previous head 4f93aa233e: 4 failed/77 passed; current head: 81 passed. Initial diagnostic-loss regression also failed on unchanged main.
  • Independent initial review: no regression findings. Cubic's classification finding fixed and resolved.
  • All required CI checks passed at this head, including Coverage, Integration Tests, Android, Linux, and Apple attempt 2 (iOS, host cleanup, and macOS).
  • Infrastructure retries: Android's first SDK archive installation failed; Apple's first host cleanup failed with a MobileCal crash dialog obscuring the hosted-network prompt. Both passed their single retry. iOS and the new probe traces passed in both Apple attempts.
  • No device behavior changed. Original hosted probe stall's cause remains unproven.