Improve coverage failure diagnostics

[XuPeng-SH]

What Changed

• Add scripts/summarize_ut_report.py to summarize go test -json reports on UT failure.
• Run the UT summary only after UT failure, as a best-effort diagnostic step (continue-on-error: true).
• Append the bounded UT failure summary to GITHUB_STEP_SUMMARY and save it as ut-failure-summary.md.
• Upload UT-Report.out, ut-failure-summary.md, and ut-report with coverage failure artifacts.
• Make coverage cleanup tolerant of no running mo-service by using pgrep ... || true.
• Avoid masking the original UT/BVT failure when raw coverage or BVT covdata is unavailable after an earlier failed/skipped step.

Why

Coverage failures can currently end with low-signal cleanup or artifact-generation errors, while the actionable go test -json failure is only available after manually downloading and parsing large artifacts. This makes failures like Go vet/build failures hard to diagnose from the CI page.

Safety

• Successful UT output remains unchanged: the go test command still redirects JSON to UT-Report.out, and the summary step does not run.
• The existing later Checkout for Python file step remains in place for the normal coverage parser path.
• Diagnostic output is bounded by default:
  • max 10 build failures
  • max 10 failed test/package events
  • max 80 output lines per failure
  • max 20 panic-like samples
• The summary script scans the report twice and keeps only bounded tails for failed packages/tests, avoiding whole-report output buffering.

Validation

• Synthetic build-fail report: summary prints the frontend vet failure clearly.
• Synthetic many-fail report: summary truncates and prints the omitted count.
• python3 -m py_compile scripts/summarize_ut_report.py
• YAML parses with PyYAML.
• actionlint passed after filtering existing custom-runner/existing-expression warnings:
  • custom runner labels: arm64-mo-shanghai-4c8g, amd64-mo-guangzhou-2xlarge16
  • existing needs.setup_mo_test_env expression warnings