- Harden LLM relation extraction result handling to parse instructor/OpenAI/Groq variations
- Add structured JSON fallback when typed generation yields zero relations
- Strip acceptance of extra kwargs like max_tokens/max_entities_prompt in relation extraction internals
- Add comprehensive unit tests with mocked LLM provider
- Add integration tests for Groq provider with environment variable API key
- Ensure relation extraction completes and returns results when model identifies relations
