Stabilize candidate-ordered recovery test

[Widthdom]

Summary

• Replace the candidate-ordered serial recovery test's narrow five-second wall-clock assertion with synchronization-state assertions.
• Preserve regression sensitivity by proving that the fatal result returns while the earlier contract candidate is still incomplete and the parallel workers have not stopped.

Validation

• dotnet build CodeIndex.sln -c Release -p:UseSharedCompilation=false
• Focused test on net8.0 and net9.0: 1 passed per target
• dotnet test CodeIndex.sln -c Release -p:UseSharedCompilation=false --no-build --no-restore
  • net8.0: 11,246 passed, 7 skipped, 0 failed
  • net9.0: 10,779 passed, 413 skipped, 0 failed
• dotnet format CodeIndex.sln --verify-no-changes --no-restore
• dotnet run --project tools/CodeIndex.Changelog -- check
• Codex adversarial review: No blocking/actionable issues found.

Documentation and changelog

• Updated the English and Japanese sections of TESTING_GUIDE.md.
• Added changelog.d/unreleased/5063.internal.md.

Follow-up candidates

• None.

Fixes #5063