Capture the model used to make OpenAI API requests

We apply additional instrumentation to gather prompts, completions and token usage, but don't have the model used to make the request, yet.

It is stored as a private attribute on the `OpenAIResponsesModel` class, so it's hard to access.

It is possible we will need to move our instrumentation up to the OpenAI client level to extract all of the default parameters. We should be able to get model params (temperature, top_p, etc) from there, too.