feat(safety): adding safety hooks

[manas-fortifyroot]

Summary

• add a minimal FortifyRoot safety hook registry in a new opentelemetry-instrumentation-fortifyroot package
• wire prompt/completion safety into the first provider wave:
  • OpenAI
  • Anthropic
  • Cohere
  • Google Generative AI
• keep upstream-facing diffs in existing provider files narrow by moving most new logic into provider-local safety.py helper modules
• emit event-only fortifyroot.safety.violation telemetry with finding-local fields only
• keep current scope to prompt enforcement plus non-stream completion enforcement
• add safety-focused tests for the new hook paths and helper logic
• add a PR GitHub Actions workflow that runs ./scripts/run-tests.sh --safety for any pull request target branch
• add a repo-level scripts/run-tests.sh helper with shared .venv support and safety-only mode

Scope Notes

• current provider slice covers OpenAI, Anthropic, Cohere, and Google Generative AI
• streaming completion enforcement is still deferred
• remaining provider parity and LiteLLM work are not part of this PR

Testing

• ./scripts/run-tests.sh --safety