### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

An A2A agent is registered on the proxy with `agent_card_params.protocolVersion: "1.0"`. The agent is built on `a2a-sdk` 1.x and implements only the protocol-1.0 JSON-RPC methods (`SendMessage`, `SendStreamingMessage`, protobuf-JSON envelopes).

Invoking this agent through the native `/a2a/{agent_id}` route works, since LiteLLM normalizes the wire format to the pinned version as documented in [Protocol versioning](https://docs.litellm.ai/docs/a2a#protocol-versioning).

Invoking the **same agent** through the documented OpenAI-compatible bridge (`model: "a2a/<agent-name>"` on `POST /v1/chat/completions`, [docs: Invoking A2A Agents](https://docs.litellm.ai/docs/a2a_invoking_agents)) always fails:

``` InternalServerError: Error code: 500 - {'error': {'message': 'litellm.APIConnectionError: A2aException - A2A error: Method not found', 'type': None, 'param': None, 'code': '500'}} ```

The bridge sends legacy 0.3-style JSON-RPC directly to the upstream agent URL:

```json { "jsonrpc": "2.0", "id": "...", "method": "message/send", "params": {…