fix(medhelm): use quasi match for PubMedQA

[MukundaKatta]

Summary

• switch PubMedQA to quasi_exact_match so punctuation-only differences do not count as wrong answers
• update the checked-in MedHELM schema metadata to keep the leaderboard config in sync
• add a regression test for the scenario main metric

Testing

• python3 -m compileall src/helm/benchmark/scenarios/pubmed_qa_scenario.py src/helm/benchmark/scenarios/test_pubmed_qa_scenario.py
• PYTHONPATH=src python3 -m pytest /tmp/helm-user/src/helm/benchmark/scenarios/test_pubmed_qa_scenario.py (fails locally: ModuleNotFoundError: No module named cattrs from repo test setup)