LightRAG / RAGAnything separation

The new FastAPI AI Integration uses RAGAnything to implement RAG with its AI features. LightRAG provides a separate web server, and at the conclusion of my thesis, RAGAnything (the multimodal RAG framework based on LightRAG) appears to be integrated with this service.

Instead of importing RAGAnything / LightRAG within the FastAPI backend with Python dependencies, the instance should be migrated to a separate Docker container. This will give greater autonomy.

Essential for this migration is that the service supports workspace separation, as each tour and some caches use different workspaces, so that the results do not get mixed up.