Enable Custom LLMs to use Tools

[nweatherburn]

Description

This PR enables Custom LLMs to yield ChatMessage's as well as strings in customStreamChat. This enables customLLMs to use tools as well. The change is backwards compatible.

Checklist

• The relevant docs, if any, have been updated or created
• The relevant tests, if any, have been updated or created

Screenshots

N/A

Testing instructions

After launching the application in debug mode, you can edit your config.ts to the below to see both messages getting correctly yielded.

export function modifyConfig(config: Config): Config {
  config.models = [
    {
      options: {
        model: "Example Model",
        title: "Example Title",    // Required for ModelDescription
      },
      streamChat: async function* (
        messages: ChatMessage[],
        signal: AbortSignal,
        options: CompletionOptions,
        fetch: (input: RequestInfo | URL, init?: RequestInit) => Promise<Response>,
      ): AsyncGenerator<ChatMessage | string> {
        yield "This is a string";
        yield {role: "assistant", content: "This is a ChatMessage"};
      }
    }
  ];

  return config;
}```

[netlify]

✅ Deploy Preview for continuedev canceled.

Name	Link
🔨 Latest commit	14e30d8
🔍 Latest deploy log	https://app.netlify.com/sites/continuedev/deploys/67c8d8a2cbcabe0008be894e

[sestinj]

thanks @nweatherburn this is great!