Check for cached tokens returned by API

[ppinchuk]

OpenAI now returns information about cached tokens, which can help make our cost estimates more accurate. Implement a check for this output and have it influence the cost computation.