fix(agent): handle non-dict LLM response gracefully and delay setting task to FAILED during Celery retries

clean up celery_worker.py by removing rate-limit delays and logs for paid Gemini key

optimize search result limit to 2 and sleep delay to 12s for OpenRouter free tier stability

add 4-second rate-limiting delay between hypothesis evaluations to respect free tier limits

integrate ResearchAgent loop and update_hypothesis database updates in Celery worker

fix: enforce PROMETHEUS_MULTIPROC_DIR at metrics layer top of file to fix multiprocess isolation

feat: configure prometheus multiprocess metrics collection for celery and fastapi

feat: add prometheus metric tracking for token counts, agent latency, and failure rates

feat: setup celery app initialization and async run_case_manager_task task

feat: added files and folders