feat(benchmarks): add PRBench

[Yunnglin]

Summary

• add the ScaleAI/PRBench benchmark with all four official Finance and Legal splits
• reproduce the official multi-turn prompt construction, reference-text handling, rubric judge contract, and weighted scoring formulas
• report clipped_score:clipped_mean as the primary metric and normalized_score:mean independently, without a cross-split OVERALL row because hard splits overlap full splits
• add generated metadata/docs, metric semantics, benchmark tests, and judge fail-closed coverage

Validation

• make lint
• python -m pytest tests/benchmark/test_prbench.py tests/api/judge/test_migrated_adapters.py tests/api/judge/test_gates.py -q (34 passed)
• python -m pytest tests/cli/test_metric_semantics_e2e.py -q -p no:benchmark (30 passed)
• python -m pytest tests/cli/test_all.py::TestRun::test_ci_lite -q -p no:warnings (passed)
• one-sample real API E2E: 18/18 rubric verdicts parsed; independently recomputed scores matched the report

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.