From RAG to Agentic: Validating Islamic-Medicine Responses with LLM Agents