Towards Omni-RAG: Comprehensive Retrieval-Augmented Generation for Large Language Models in Medical Applications