Feature request

Add actual information about the use of the KV cache when processing a request, which is assumed in completion_usage in the OpenAI API.

Describe the problem you're encountering

Almost all providers provide discounts for cached tokens, so the lack of support may be a product blocker of launching on top of Dynamo.

In addition, actual per-request KV cache usage data will help to compare different caching improvements (KV-aware-router, KVBM, ...)

Describe alternatives you've tried

No response