Add round-robin provider rotation across 9 API keys

Batch emotion + profile extraction into single LLM call per coaching turn

Add automatic provider fallback on rate limiting

Support GROQ_API_KEY env var and strip whitespace from config

Make LLM provider configurable via env vars (switch to Groq)

Fix emotion detection: center arousal in backend, fix LLM classifier fallback, add 57 emotion accuracy tests

Add SSE streaming for token-by-token LLM responses and fix circumplex display

Fix double cognitive load assessment causing LLM repetition

Improve conversation naturalness and reduce number recitation

Fix duplicate user message bug causing LLM repetition, add debug logging

Fix LLM response rambling and topic hijacking

Initial commit: MindSphere Coach — ToM-powered coaching agent