### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### The Feature

LiteLLM should preserve Anthropic's prompt cache creation TTL breakdown in OpenAI-compatible Chat Completions usage responses.

Anthropic returns cache write token details under `usage.cache_creation`, for example:

```json { "cache_creation": { "ephemeral_5m_input_tokens": 0, "ephemeral_1h_input_tokens": 6179 } } ```

LiteLLM currently exposes the aggregate `cache_creation_input_tokens`, but downstream clients cannot reliably distinguish how many cache write tokens used the 5 minute TTL versus the 1 hour TTL in Chat Completions responses, especially for streaming responses where usage is aggregated before being returned to the client.

The requested behavior is to include the TTL-level breakdown in Chat Completions usage while keeping the existing aggregate fields:

```json { "usage": { "cache_creation_input_tokens": 6179, "cache_read_input_tokens": 0, "cache_creation": { "ephemeral_5m_input_tokens": 0, "ephemeral_1h_input_tokens": 6179 } } } ```

### Motivation, pitch

This is needed for accurate down…