feat(01-02): LLM Backend Pool — LiteLLM Router with Ollama + Anthropic + OpenAI fallback

- Create llm_pool/router.py: LiteLLM Router with fast (Ollama) and quality (Anthropic/OpenAI) model groups
- Configure fallback chain: quality providers fail -> fast group
- Pin LiteLLM to ==1.82.5 (avoid September 2025 OOM regression in later releases)
- Create llm_pool/main.py: FastAPI service on port 8004 with /complete and /health endpoints
- Add providers/__init__.py: reserved for future per-provider customization
- Update docker-compose.yml: add llm-pool and celery-worker service stubs

packages/llm-pool/llm_pool/__init__.py

@@ -0,0 +1,10 @@
+"""
+konstruct-llm-pool — LiteLLM Router service.
+
+Exposes a FastAPI HTTP API for LLM completions with multi-provider routing
+and automatic fallback. Import the FastAPI app from main.py.
+"""
+
+from llm_pool.router import complete, llm_router
+
+__all__ = ["complete", "llm_router"]