Stabilize parallel snapshot refresh test

[Widthdom]

Summary

• give each signal-gated status wave in the checkpointed-WAL parallel batch regression test two minutes of net9 full-suite headroom
• preserve the existing completion signals, refreshed-result assertions, and exact session-count assertion
• document the timeout contract in both TESTING_GUIDE languages and add a bilingual changelog fragment

Root cause

The regression test used independent 60-second bounds for each status wave. Under a loaded net9 full-suite run, the first bound could be exhausted even though the same test completed successfully in isolation, and the 15-second cancellation cleanup made the reported failure appear at roughly 75 seconds.

Impact

Test-only stabilization. Production behavior is unchanged, and the concurrency contract remains fully asserted.

Validation

• Release solution build: 0 warnings, 0 errors
• targeted net8.0 test: passed
• targeted net9.0 test: passed
• net9.0 full suite: 10,622 passed, 420 skipped, 0 failed (11,042 total; 1 h 6 m)
• changelog fragment validation: 105 fragments passed
• dotnet format CodeIndex.sln --verify-no-changes --no-restore
• git diff --check origin/main..HEAD
• Codex adversarial review: no blocking/actionable issues found

Documentation and changelog

• updated the English and Japanese TESTING_GUIDE.md entries
• added changelog.d/unreleased/5015.fixed.md

Follow-up candidates

None.

Fixes #5015