Add new prompt caching options and detailed diagnostics

Introduce prompt cache control options for GPT-5.6+ models to specify caching behavior and diagnostics. Add detailed prompt cache diagnostics models to provide insights on cache hits, misses, and reasons for misses. Update relevant response schemas to include these prompt cache options and diagnostics. Additionally, add new error responses for inference rate limits and temporary service unavailability with retry headers for better handling of throttling and overloads.

OpenAI-API-Ref-Source-Revision: 1051358