Add an option to override the base_url

[intellectronica]

Adds an option -o base_url https://... to override the endpoint used for Command-R.

This is useful when using the model through a cloud provider or your own deployment (tested this with a deployment on Azure - after deploying Command-R+ through the Azure AI Studio I have an endpoint: https://....swedencentral.inference.ai.azure.com/v1).

Ultimately I think this is something that can be managed by llm in a standard way, like keys, since most popular models are now available from various cloud providers or as your own deployment (if the weights are available). If you think this is interesting let me know and I don't mind contributing that.

[simonw]

This one feels a bit weird to me as a model option, mainly because none of the other LLM plugins have used options in this way.

I'm going to implement this as an environment variable instead.