Problem

When using Azure OpenAI as a model provider, all potentially available models from this service are shown, but that likely doesn't correspond to the deployed models in a workspace of this service--we have 4 deployed models. You of course get an error if you try to call a model that is not deployed.

Solution

The fix in this PR adds a flag that allows filtering to only the models defined in the config file.

Example
Here, no models will be shown for Copilot despite a Github token being found, and only 3 models of 4 deployed are shown, with the rest found and added to the database being hidden.

{
  "$schema": "https://opencode.ai/config.json",
  "onlyConfigModels": true,
  "provider": {
    "github-copilot": {
      "models": {}
    },
    "azure": {
      "models": {
        "GPT-4.1": {
          "name": "GPT4.1"
        },
        "GPT-4.1-mini": {
          "name": "GPT4.1-mini"
        },
        "GPT-4.1-nano": {
          "name": "GPT4.1-nano"
        }
      }
    }
  }
}

Alternatively, this could just be default behavior omitting the need for such a flag.

It would also be nice to query the available models to prevent the need to define models in the config., but the included solution is easy enough and allows some additional flexibility about which models to show.