Return prompt cache utilization on completion responses

[reckart]

Since Ollama has prompt caching now (right?), it would be great if the utilization of the cache could be returned in requests.

E.g. the OpenAI-compatible API could be extended with the new usage/prompt_tokens_details/cached_tokens.

A similar field in the Ollama API would also be great.

[codefromthecrypt]

been a while on this and the lack of cached_tokens and reasoning_tokens interferes with openai-agents SDK usage. will try to get a workaround there until this is sorted here

[codefromthecrypt]

raised this but not sure when it will merge.. openai/openai-agents-python#2034