Add benchmark comparison workflow with PR comments

- benchmark.yaml: on PRs, runs pytest-benchmark on both main and PR,
  compares results via tests/benchmark/compare.py. On tags, saves
  benchmark data permanently as artifacts (retention-days: 0).
- benchmark_comment.yaml: posts the comparison as a PR comment
  (hides old comments, same pattern as primer_comment.yaml)
- tests/benchmark/compare.py: reads two pytest-benchmark JSON files,
  computes median deltas, outputs markdown with ±5% threshold flags

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>