`ChatDeepSeek` fails with models that return `reasoning_content` (e.g. `deepseek-v4-flash`) when tools are used

[Damonsoul]

Description:

When using ChatDeepSeek with a model that outputs a reasoning_content field (like deepseek-v4-flash), calling a tool causes a 400 error from the API because the reasoning_content is not passed back in the next request.

Steps to reproduce:

import os
from chatlas import ChatDeepSeek
import requests

chat = ChatDeepSeek(model='deepseek-v4-flash')

def get_current_weather(lat: float, lng: float):
    """
    Get the current weather given a latitude and longitude.

    Parameters
    ----------
    lat: The latitude of the location.
    lng: The longitude of the location.
    """
    lat_lng = f"latitude={lat}&longitude={lng}"
    url = f"https://api.open-meteo.com/v1/forecast?{lat_lng}&current=temperature_2m,wind_speed_10m&hourly=temperature_2m,relative_humidity_2m,wind_speed_10m"
    response = requests.get(url)
    return response.json()["current"]

chat.register_tool(get_current_weather)
chat.chat("How's the weather in San Francisco?")

Traceback:

BadRequestError: Error code: 400 - {'error': {'message': 'The `reasoning_content` in the thinking mode must be passed back to the API.', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_request_error'}}

Expected behaviour:
When the assistant message contains a reasoning_content field (as part of the model’s thinking mode), chatlas should include it in the subsequent API call after a tool result is received. Without it, the DeepSeek API rejects the request.

Environment:

• chatlas version: 0.15.2
• Python version: 3.13.3
• Model: deepseek-v4-flash

Additional context:
This issue seems to affect any DeepSeek model with thinking mode enabled that requires reasoning_content to be echoed back exactly as received. Many new models (including deepseek-v4-pro and deepseek-v4-flash) behave this way. It would be great if chatlas could automatically preserve and return the reasoning_content during multi-turn interactions, especially tool use.

[cpsievert]

This particular example is fixed by #295. See #288 (comment) if you're still having issues with more complex multi-turn scenarios.