add automated flaky test detection to pr-status script
Checks last 5 failed CI runs across all branches (excluding current PR).
A test is flagged as flaky if it fails on 2+ different branches. Skips
runs with >20 failed jobs (systemic failures). Parallelizes API calls
for speed. Results shown in index.md "Known Flaky Tests" section and
written to flaky-tests.json. Use --skip-flaky-check to skip.
Also updates pr-status skill with CI repro best practices: always match
exact env vars when reproducing locally.
