assistant: Make it easier to define custom models (#15442)

This PR makes it easier to specify custom models for the Google, OpenAI,
and Anthropic provider:

Before (google):

```json
{
  "language_models": {
    "google": {
      "available_models": [
        {
          "custom": {
            "name": "my-custom-google-model",
            "max_tokens": 12345
          }
        }
      ]
    }
  }
}
```

After (google):

```json
{
  "language_models": {
    "google": {
      "available_models": [
        {
          "name": "my-custom-google-model",
          "max_tokens": 12345
        }
      ]
    }
  }
}
```

Before (anthropic):

```json
{
  "language_models": {
    "anthropic": {
      "available_models": [
        {
          "custom": {
            "name": "my-custom-anthropic-model",
            "max_tokens": 12345
          }
        }
      ]
    }
  }
}
```

After (anthropic):

```json
{
  "language_models": {
    "anthropic": {
      "version": "1",
      "available_models": [
        {
          "name": "my-custom-anthropic-model",
          "max_tokens": 12345
        }
      ]
    }
  }
}

```

The settings will be auto-upgraded so the old versions will continue to
work (except for Google since that one has not been released).

/cc @as-cii 

Release Notes:

- N/A

---------

Co-authored-by: Thorsten <thorsten@zed.dev>

crates/language_model/src/language_model.rs

@@ -13,14 +13,15 @@ use futures::{future::BoxFuture, stream::BoxStream};
 use gpui::{AnyView, AppContext, AsyncAppContext, SharedString, Task, WindowContext};
 
 pub use model::*;
+use project::Fs;
 pub use registry::*;
 pub use request::*;
 pub use role::*;
 use schemars::JsonSchema;
 use serde::de::DeserializeOwned;
 
-pub fn init(client: Arc<Client>, cx: &mut AppContext) {
-    settings::init(cx);
+pub fn init(client: Arc<Client>, fs: Arc<dyn Fs>, cx: &mut AppContext) {
+    settings::init(fs, cx);
     registry::init(client, cx);
 }