## What happened?

When using LiteLLM proxy with the **DeepSeek anthropic-compatible endpoint** (`api_base: https://api.deepseek.com/anthropic`, served via `DeepSeekAnthropicMessagesConfig`), any multi-turn thinking-mode conversation whose assistant history is missing `reasoning_content` fails with HTTP 400:

``` litellm.BadRequestError: DeepseekException - {"error":{"message":"The `reasoning_content` in the thinking mode must be passed back to the API.","type":"invalid_request_error","param":null,"code":"invalid_request_error"}} ```

This is the same error message tracked in #26395 / #27946 / #28045, but **on a different code path that none of those issues or their fixes address**:

- **#26395** locates the root cause in `litellm/types/utils.py` (`Message.__init__` strips `reasoning_content` when `None` during serialization) — a **prevention** fix that preserves `reasoning_content` from the response. - **#27946** locates the root cause in `litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py` (Anthropic → OpenAI Chat Completions adapter drops `reasoning_content`) — only fires when `LITELLM_USE_CHAT_COMPLETIONS_URL_FOR_ANTHROPIC_MESSAGES=true`. - **This issue…