Add generic LLM API integration for chat review

[kerbearasaurus]

Summary

• Adds optional LLM API configuration (base URL, API key, header, model) to settings
• Users can configure any OpenAI-compatible chat completions API (EvidenceMD, OpenAI, Anthropic, etc.)
• After PhenoML agent responds, users can click "Get LLM Review" to send response to configured LLM
• LLM response appears as a labeled chat message, allowing comparison with agent output

Implementation

• New src/lib/llm.ts: Generic sendChatCompletion() function supporting any chat API
• Settings form updated with LLM config fields (base URL, API key, key header, model)
• Chat pipeline adds sendToLlm() function and ADD_ASSISTANT_MESSAGE action
• Chat messages enhanced with optional source field to label LLM responses
• Button only shows when LLM is configured and after agent response completes

Test plan

• Open settings and verify LLM API fields appear (optional)
• Save LLM config with test credentials
• Run normal PhenoML pipeline to completion
• Verify "Get LLM Review" button appears on agent response
• Click button and verify LLM response appears with "LLM Review" badge
• Verify button does NOT appear when LLM API not configured

🤖 Generated with Claude Code

[kerbearasaurus]

Just making this temp PR for the hackathon