[Feature request] Support for LiteLLM and LiteLLM proxy

[danrr]

LiteLLM (https://github.com/BerriAI/litellm) provides a common interface (using the OpenAI API) for a lot of models and providers.

Has support for LiteLLM been considered?

I can imagine it as a prefix for the model given as input e.g. litellm/azureai/llama-2-70b-chat-wnsnw for the litellm package and litellm-proxy/azureai/llama-2-70b-chat-wnsnw for the proxy using openai.

This could help expand the range of models and features beyond what's available in inspect_ai without requiring a custom model implementation from the consumer.