feat(metrics): add JSON schema conformance metric

[aayushi-sing]

Summary

Adds a deterministic JSON Schema conformance metric for evaluating structured model outputs.

Changes

• Add SchemaConformance generation metric.
• Parse generated answers as JSON and validate them against JSON Schema Draft 2020-12.
• Return 1.0 for fully conformant outputs and 0.0 for invalid JSON or non-conformant outputs.
• Add deterministic top-level field-level partial scoring for object schemas.
• Pass dataset metadata through the evaluation pipeline so metrics can access the expected JSON schema.
• Register schema_conformance in METRIC_REGISTRY.
• Add jsonschema dependency and update uv.lock.
• Add unit tests covering valid, invalid, partial, missing-schema, empty-schema, and extra-property cases.

Testing

• uv run pytest tests/unit/test_metrics/test_generation.py -q — 41 passed
• uv run ruff check openagent_eval/metrics/generation/schema_conformance.py tests/unit/test_metrics/test_generation.py — passed
• uv run ruff format --check openagent_eval/metrics/generation/schema_conformance.py tests/unit/test_metrics/test_generation.py — passed
• git diff --check — passed

The full test suite was also run locally. It had one unrelated Windows-specific failure in test_oaeval_help_runs_without_pytest caused by a cp1252 Unicode encoding error; no changes were made to that unrelated test.
Closes #347