MCP sampling/createMessage response includes system prompt text in content

### Describe the bug

When an MCP server sends a `sampling/createMessage` request, the Copilot CLI correctly routes it to the LLM and returns a response. However, the response content prepends the entire system prompt text before the LLM's actual answer.

### Affected version

GitHub Copilot CLI 1.0.16-0

### Steps to reproduce the behavior

1. Configure an MCP server that uses `sampling/createMessage` with a system prompt
2. Have the server send a sampling request where it expects a specific response format
3. Approve the sampling request when prompted
4. Observe the response content includes the system prompt prepended to the LLM's answer

### Expected behavior

The `CreateMessageResult.Content` should contain only the LLM's response text, e.g.:

```
=== FALSE
```

### Additional context

## Actual behavior

The response content contains the system prompt echoed back, followed by the LLM's answer:

```
... === FALSE
```

## Note

This works correctly in VS Code's MCP sampling implementation — the system prompt is not included in the response content.

Start by reproducing an MCP server's sampling/createMessage request with a system prompt, then inspect the returned CreateMessageResult.Content. Compare the CLI result with VS Code's MCP sampling behavior; done means the content contains only the LLM response, without the system prompt text.

Escrita pelo modelo de indexação a partir do texto da issue.