feat: Add public safety training data pipeline

[homanp]

Summary

• Add a public safety training dataset pipeline that converts configured pairwise and single-response safety sources into the project schema.
• Add the generated dataset summary and track the large training JSONL through Git LFS.
• Update training defaults, Modal sampling, docs, and tests for the new data/training_data.jsonl flow.

Test plan

• uv run pytest
• Modal guarded generation smoke checks with the saved probe_out_public_safety probe

[open-cla]

Contributor License Agreement

All contributors are covered by a CLA.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.