chore(ci): benchmark pipeline and reintroduce caching improvements #18

Context

We are intentionally shipping macOS wheel build support first and deferring cache optimization changes to a follow-up. This issue tracks the benchmarking + A/B comparison so we can quantify impact before and after reintroducing caching improvements.

Plan

1. Merge MR !30 as the baseline (no additional cache optimizations).
2. Capture baseline timing from 3-5 comparable pipelines (same runner class and similar workload window).
3. Implement caching improvements in a separate MR.
4. Capture post-change timing from 3-5 comparable pipelines.
5. Compare medians and decide whether to keep/tune/revert cache changes.

Metrics to Collect

• Per-job median duration for:
  • build_ci_image
  • lint
  • rust_test
  • python_test
  • publish_main_artifacts / publish_tag_artifacts (for release/main runs)
• Overall pipeline median duration.

Acceptance Criteria

• Baseline and post-change measurements are documented.
• Comparison uses the same runner architecture/class.
• Follow-up MR includes before/after summary and recommendation.

Originally by @pimlock on 2026-02-17T21:22:57.862-08:00