## Description

When using the OpenAI Responses API endpoint (`/v1/responses`) with `cache_control` on `input_text` content blocks, the `cache_control` field is silently dropped during the Responses → Chat Completions transformation. This means prompt caching directives sent via the Responses API never reach the underlying provider (e.g. Anthropic/Bedrock).

This is inconsistent: `cache_control` on **tools** is correctly preserved through the same transformation.

## Reproduction (unit test — no API call needed)

```python from litellm.responses.litellm_completion_transformation.transformation import ( LiteLLMCompletionResponsesConfig, )

# Case 1: input_text with cache_control — DROPPED result = LiteLLMCompletionResponsesConfig.transform_responses_api_request_to_chat_completion_request( model="claude-sonnet-4-6", input=[{ "type": "message", "role": "user", "content": [{ "type": "input_text", "text": "Hello", "cache_control": {"type": "ephemeral", "ttl": "1h"}, }], }], responses_api_request={}, ) messages = result.get("messages", []) content = messages[-1].get("content", []) print(content[0].get…