- Update conversation evaluation templates with more realistic test cases
- Streamline safety evaluation types by removing redundant checks
- Update RAGAS evaluation metrics to remove deprecated ones
- Standardize context evaluation documentation and examples
- Add missing AgentJudge evaluation type to custom evaluations

Key changes:
- Conversation templates now use math homework example for better clarity
- Removed redundant safety checks like ProfanityFree and NotUnusualPrompt
- Removed deprecated RAGAS metrics (AnswerSemanticSimilarity, Maliciousness)
- Updated context evaluation documentation with comprehensive examples
- Added AgentJudge as new custom evaluation type