Investigate & possible fix metering for gemini models search and caching

<img width="1050" height="352" alt="Image" src="https://github.com/user-attachments/assets/e5a91c75-fe3b-4bce-821d-fd9f18b0922f" /> Newer gemini models support things like grounded searches and timed caching. However openai compatible api might not expose these costs in the input/output tokens the way claude api does. verify that metering for this is valid, and fix if not