[Langchain-Deepseek] _Get_request_payload Drops Reasoning_content From Prior Assistant Messages, Causing 400 Errors On Tool Calls

The `_get_request_payload` method (and/or its underlying message converter) builds OpenAI-style message dicts by copying only known fields like role, content, and tool_calls, ignoring custom `additional_kwargs` such as `reasoning_content`. DeepSeek's thinking mode API contract requires that `reasoning_content` from prior assistant messages be included in follow-up requests, so its omission leads to invalid_request_error 400 responses.

Run the following with langchain-deepseek 1.1.0: ```python from langchain_deepseek.chat_models import ChatDeepSeek from langchain_core.messages import AIMessage, HumanMessage model = ChatDeepSeek(model="deepseek-v4-flash", api_key="fake_api_key") messages = [ HumanMessage(content="Hello!"), AIMessage( content="Hello, can i help you?", additional_kwargs={"reasoning_content": "Simple greeting, Just greet back in English casually."} ) ] print(model._get_request_payload(messages)) ``` Observe that the assistant message dict lacks `reasoning_content`. When such messages are sent in a multi-round conversation with tool calls, the API returns `openai.BadRequestError: 400 - The 'reasoning_content' in the thinking mode must be passed back to the API.`

This is a temporary workaround that relies on the current signature and return shape of _get_request_payload. If upstream changes the method or the payload structure, the patch may need adjustment. The `not in` guard already avoids redundant injection if an official fix later adds the field natively. Users should only use reasoning-capable DeepSeek models; passing reasoning_content to non-reasoning models might cause API errors. Rollback is simply removing the monkey-patch block and upgrading to a version that includes the official fix.