A couple of use-cases for this. First, if you hit e.g. `gemini/gemini-1.5-flash-latest` and the API returns whatever that "latest" version actually is would be nice to record that. Secondly, `llm-llama-server` needs this - it _always_ records the model ID `llama-server` even though that software could be serving Mistral or Llama or Gemma or Qwen or who knows what: - https://github.com/simonw/llm-llama-server/issues/2