[Docs] Propose a "Best Practice" example for using VertexAiRagRetrieval in a multi-agent system

*Transferred from https://github.com/google/adk-samples/issues/174*

When including the `VertexAiRagRetrieval` tool in a main orchestrator agent with other custom Python function tools, a `400 INVALID_ARGUMENT` error occurs. This is due to a core API constraint that only supports multiple tools if they are all search tools.

**The Solution:**

* **Isolate the RAG Tool:** Create a dedicated sub-agent to host and expose the `VertexAiRagRetrieval` tool.
* **Delegate from the Orchestrator:** The orchestrator agent calls the dedicated sub-agent using an `AgentTool` instead of directly calling the RAG tool.

Original post says: It would be incredibly helpful for the community if the official ADK documentation or examples included a section highlighting this "tool-runner agent" pattern. It's a key piece of practical knowledge for building robust, real-world multi-agent systems with the ADK.