ci: track test flavor in per-test baseline and rename job to Track Test Duration

[hsuan-lun-chiang]

Description

This PR addresses follow-up feedback from b/528433969 by mitigating hardware-dependent test duration noise in the Per-Test Baseline comparison, excludes noisy CPU runners from macro-level dashboard alerting, and updates workflow job naming.

Problem & Context

1. Previously, per_test_baseline.json tracked test durations using un-flavored test names. When a test ran on multiple hardware platforms with different execution times (e.g., QKClipMLATest.test_mla_dot_product_integration taking 0.94s on CPU vs 15.99s on GPU), baseline comparison falsely flagged regressions when comparing across hardware types.
2. Macro-level dashboard tracking (github-action-benchmark) triggered false-positive CI alerts when total execution times on shared GitHub Actions CPU runners fluctuated (~25% variance due to multi-tenancy noise).

Solution

1. Track Hardware Test Flavor (flavor::test_name): Baseline keys in per_test_baseline.json are now formatted as flavor::test_name (e.g., gpu-unit::..., tpu-unit::..., tpu7x-unit::...), isolating durations for each hardware flavor and device type.
2. Exclude CPU Suites from Macro-Level Alerts: In process_test_results.py, CPU test suites (cpu-unit, cpu-torch-reference, cpu-post-training-unit) are skipped when generating benchmark-results.json. This prevents false-positive CI failure alerts from github-action-benchmark caused by shared CPU runner noise, while still maintaining individual CPU test baselines in per_test_baseline.json.
3. Legacy Key Clean-up: Automatically purges un-flavored legacy baseline keys when initializing new_per_test_baseline.json.
4. Rename Job & Workflow: Updated job name from Track Test Performance to Track Test Duration in track_performance.yml and ci_pipeline.yml.
5. Security & Authentication: Pinned action references to full commit SHAs (zizmor compliance) and added explicit GITHUB_TOKEN remote authentication for pushing per_test_baseline.json to gh-pages when persist-credentials: 'false' is used.
6. Unit Tests: Added tests/utils/process_test_results_test.py to verify per-flavor baseline isolation, regression checking, and CPU macro-benchmark exclusion.

Tests

Ran local unit tests and pre-commit hooks:

python3 tests/utils/process_test_results_test.py
pre-commit run --files tests/utils/process_test_results.py tests/utils/process_test_results_test.py .github/workflows/track_performance.yml .github/workflows/ci_pipeline.yml

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.