[BUG] Gateway model discovery `capabilities.image_input.supported=false` is ignored in session image behavior

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

When using gateway model discovery, Claude Code receives a model entry with:

"capabilities": {
  "image_input": {
    "supported": false
  }
}
but the session still behaves as if the model supports image input.

This makes it hard for gateway/proxy providers to correctly advertise non-vision models.

### What Should Happen?

If a model returned by gateway model discovery has:

"capabilities": {
  "image_input": {
    "supported": false
  }
}
Claude Code should treat that model as not vision-capable in session behavior.

Examples of expected behavior:
image upload should be disabled for that model, or
image content blocks should not be sent to the model, or
Claude Code should surface a clear error before sending the request

### Error Messages/Logs

```shell

Steps to Reproduce

Configure Claude Code to use a gateway model discovery endpoint
Return a model from /v1/models like:
{
"id": "glm-5.1",
"capabilities": {
"image_input": {
"supported": false
}
},
"created_at": "2026-05-26T06:17:26Z",
"display_name": "andes-glm-5.1",
"type": "model"
}
Start a session with that model
Try using image input in the session

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

v2.1.150

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_