support seperate thinking process in the stream chunk

[arvinxx]

We are intergating the ollama with DeepSeek R1, and we find current Ollama implement is to add CoT in the content. But with official implement, it should be in another field param like reasoning_content.

as for the gemini-flash-thinking model ,they also support a seperate field to get the thinking content: https://ai.google.dev/gemini-api/docs/thinking#work-thoughts

[ecker00]

In the mean time a cheap workaround:

const [think, message] =  res.response.split('</think>').map((s) => s.trim());