When using deepseek-v4 in rimtalk, force disable its thought process. This only applies to deepseek-v4-xxx models; other models will still follow the original logic, allowing the AI to output quickly.
