feat(evaluators): Add example for LLMAJ evaluator

[wrisa]

Summary

• Added a runnable example for the galileo.llm (LLM-as-judge) evaluator, mirroring the existing examples/galileo_luna/ example. Demonstrates three control patterns: pre-execution prompt injection detection, composite post-execution toxic output scoring, and regex-based API key leak prevention.

Scope

• User-facing/API changes: None — example only.
• Internal changes: New examples/galileo_llm/ directory with setup_controls.py, demo_agent.py, pyproject.toml, and README.md.
• Out of scope: No evaluator logic changes; no SDK or server changes.

Risk and Rollout

• Risk level: Low — example code only, no production path affected.
• Rollback plan: Delete the examples/galileo_llm/ directory.

Testing

• Added or updated automated tests - N/A, example code.
• Ran make check (or explained why not) - lint passes (ruff check clean).
• Manually verified behavior - ran setup_controls.py and demo_agent.py against local server with PyCharm debugger; breakpoints hit through LlmEvaluator._evaluate and server POST /api/v1/evaluation endpoint.

Checklist

• Linked issue/spec (if applicable) - SAO-16858
• Updated docs/examples for user-facing changes - README.md documents all env vars, scorer configuration, boolean threshold pattern (operator=eq, threshold=True), and CA file setup.
• Included any required follow-up tasks - example will need updating after feature/SAO-17516-request-payload-contract merges if the control payload contract changes affect setup_controls.py.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!