chore: optimize test output for debugging
Changes to see actual test failures:
- Reduce workers from 12 to 4 (less parallelism = clearer output)
- Remove --reruns to fail fast and see real errors
- Add --tb=short to show concise tracebacks immediately
- Add --no-header to reduce noise
- Change -vv to -v (less verbose but clearer with xdist)
This sacrifices speed for visibility during active debugging.
Once tests are stable, revert to 12 workers with reruns.
