Document Storage Strategy for RAG: Unified with Vector DB or Isolated?

Overview:
Seeking best practices for RAG document chunk storage: store chunks within the vector database or use separate storage? How do you manage metadata and scaling?