feat(evaluator): bundle metrics for plugin execution

[SandyChapman]

Summary

• add MetricBundle and cloudpickle bundling primitives to the evaluator SDK
• make evaluator plugin backend bundle runtime metrics before remote submission
• hydrate bundles in plugin execution so the backend has no hard-coded metric type knowledge
• publish remote evaluator aggregate, row-score, and artifact results through Jobs result storage
• add local and remote smoke scripts covering built-in, custom protocol, and multi-metric execution

Validation

• uv run pre-commit run -a
• uv run --frozen pytest plugins/nemo-evaluator/tests/test_evaluate_job.py plugins/nemo-evaluator/tests/test_sdk_job_resources.py -q
• uv run --frozen scripts/run_evaluator_local_bundle_path.py
• uv run --frozen scripts/run_evaluator_remote_bundle_path.py

[github-actions]

Suite	Lines Covered	Line Rate	Branch Rate
Unit Tests	18392/24362	75.5%	61.8%
Integration Tests	11830/23144	51.1%	26.2%

[SandyChapman]

Closing this older draft in favor of #46, which carries the resolver-based metric bundling work.