We need a way to show an agent what a surface is displaying

In order for the user to query the agent about things they see (e.g. "Tell me about the one on the left"), the agent needs to be able to understand the layout and content of the surface.

The agent has no way to query this at the moment, and although we do have a way to send the current data model with every A2A message, we don't have any way to do the same with the UI tree.

One proposal might be to introduce a new A2UI message `querySurface` that allows the agent to get the current state of the surface, both the UI tree and the data model.

The issue proposes a new A2UI message type `querySurface`. Start by examining the A2A and A2UI message handling code to understand the protocol. Look for existing message definitions and how the data model is sent. Determine where the UI tree is constructed and serialized. Implementing this requires understanding the surface state management and how to expose it via a new message.

索引モデルが issue の本文から書いたものです。