[QA] ✅ VERIFIED - All acceptance criteria met. Server starts successfully, /health returns 200 with correct JSON structure, all 77 tests pass. Model preloading functionality works as designed. Detailed report available at /tmp/qa-LCA-8.md

Implemented FastAPI main app with the following changes: **Files created:** - src/librechat_audio/main.py - FastAPI app with lifespan context manager for model loading - src/librechat_audio/routes.py - Routes module with /health endpoint - tests/test_main.py - Tests for the app and health endpoint **Files modified:** - pyproject.toml - Added fastapi, uvicorn, pydantic dependencies - Makefile - Added make run command to start the server **Verification:** - make run starts the server on port 8000 - /health endpoint returns 200 with JSON: {status:healthy,version:0.1.0,models_loaded:[]} - All 77 tests pass (5 skipped for local model tests) **Features:** - Lifespan manages optional local STT model preloading via LIBRECHAT_AUDIO_PRELOAD_LOCAL_STT env var - Health endpoint shows loaded models for monitoring