Calling custom agent via prompt ignores configured model in agent.md

### Describe the bug

When defining a custom agent and setting the `model` property in it's `agent.md` file, the agent is called with a completely different model than defined or selected via the `/model` command in the CLI. It looks like the model that is passed down to the agent when called interactively via a instructive prompt is `gpt-4.1` and sometime `gpt-4o-mini`.

If you call the agent directly from the command prompt via the command `copilot --agent=TestAgent --prompt="Hi Agent!"` it works fine and it uses the model that is defined in the markdown file as seen below.

### Affected version

GitHub Copilot CLI 1.0.35.

### Steps to reproduce the behavior

1. Create an custom agent with the `model` property set to a model (e.g. `gpt-5.3-codex`) and instructions to expose it's model that is used.
2. Save the `xxx.agent.md` in your system-wide `.copilot\agents` folder
3. Run `copilot` in the terminal
4. Type `/model` and select a different model than the one defined in the agents markdown
5. Use a prompt to call the which you created in step 1
6. Check it's output in the CLI

### Expected behavior

The agent that is called interactively via a prompt, should use and run with the model defined in the agent markdown.

### Additional context

- OS: Windows 11 Enterprise
- PowerShell 7.6.1
- Visual Studio Code 1.117.0

Reproduce the difference between calling a custom agent through an interactive prompt after using `/model` and calling it with `copilot --agent=TestAgent --prompt="Hi Agent!"`. Start by tracing model selection for the prompt and agent invocation paths, using the `model` property in `xxx.agent.md` as the expected source. Done means both paths use the model configured in the agent file.

Do mô hình lập chỉ mục viết ra từ nội dung của issue.