docs(ci): record PR validation cost and tier ownership

[SunSi12138]

Closes #374

Summary

• records representative fresh PR Quick wall-clock runs plus a separately scoped successful quick step-timing sample
• identifies the major serial and codec-compatibility cost contributors
• records observed repeated Unit Tests flake exposure on #2484 rather than classifying it as low
• assigns every current validation check to Fast, Extended, or Nightly/merge-to-dev with rationale
• preserves the current unfiltered pull_request coverage for Fast/Extended across dev, main, and stacked feature-to-feature PRs
• defines a Fast target of <=3 minutes active time with a 5-minute hard timeout
• documents current dev/main ruleset and required-status implications

Scope

Documentation only. No workflow, trigger, required-status, or ruleset migration is included.

Baseline

Representative fresh successful PR Quick attempt-1 runs are 6m47s (#388), 7m03s (#392), and 7m06s (#412), with a 7m03s median.

#2484 attempt 3 is not treated as a fresh full-workflow wall-clock sample: it reran quick after two same-head Unit Tests failures while successful codec jobs retained results/timestamps from the earlier invocation. It is used only for detailed successful quick step timings.

PR-target coverage contract

Current PR Quick uses an unfiltered pull_request: trigger. #395 shows that a stacked PR targeting sync/386-merge-base still ran PR Quick #2416, and #390 shows that a PR targeting main ran both PR Quick #2307 and Release Gate #441.

Accordingly, #375/#376 must keep Fast/Extended on all PR targets rather than narrowing them to dev. Release Gate remains additive on main; it is not a strict superset of PR Quick (for example it lacks formatting, Debug build, Demo Oneway, and Load Smoke). Nightly/merge-to-dev remains additive post-merge coverage rather than a replacement for non-dev PR validation.

Follow-up implementation remains #375 -> #376.