Fix Phase 2 timeout: cap LLM retries, move server wait to main, add global time guard

Fix Phase 2: add server readiness check and wrap all httpx calls in try/except

Fix inference.py to match sample script exactly: HF_TOKEN var name, no fallback

Fix Phase 2: add [START]/[STEP]/[END] structured output + simulation fixes

Increase max_tokens to 4000 for thinking model compatibility

Add baseline results, outputs dir, and backup Groq fallback

Improve inference script robustness and fallback strategy

Fix HTTP endpoint statefulness and response format

Add baseline inference script and Dockerfile