Wrong model billed

### Describe the bug

When I prompted Copilot using auto model selection, it informed me that it used GPT-5.3-Codex in chat window:

But when done, it reported GPT-5.4:

Also when I opened my usage page in GitHub account settings, I saw GPT-5.4 as the only "Auto" model usage:

The thing is that the response style seemed to be that of GPT-5.3-Codex and I've been charged for GPT-5.4

### Affected version

VSCode Insiders extension - GitHub Copilot Chat v0.56.2026070103

### Steps to reproduce the behavior

1. Select "auto" model selection
2. Send prompt
3. Observe initial information on selected model
4. Observe response style
5. Observe summary information on model used
6. Observe AI usage billing chart in account settings

### Expected behavior

Initial selected model info, summary model info, actually used model and billing info should all be unanimous

### Additional context

_No response_

未指定源文件、测试或入口点。首先使用自动模型选择重现这些步骤，并比较初始模型标签、响应摘要和账户使用量图表；当这些信息始终能识别实际使用的模型，且计费信息一致时，此 issue 即完成。

由索引模型根据 Issue 内容生成。