Add support for OpenAI's Predicted Outputs

[aretrace]

OpenAI released a new api feature called Predicted Outputs. It's designed to reduce latency when most of the output is already known in advance. This is especially useful for tasks that involve minor changes to existing content (maybe like structured data transformations 😁).

[karljuhlthefool]

Upvote. Although function calling isn't supported with predicted outputs, the json mode should be implemented.

[jxnl]

we mostly pass in the arguments, does this not work by default?

[sreeprasannar]

The input should take care of itself. i.e. it should be possible to pass in the arguments all the way to the openai endpoint. From the completion response, it's useful to surface the two new fields accepted_prediction_tokens, rejected_prediction_tokens to the caller.
@jxnl Here's a PR - #1312