[Feature Request] Support streamWithThink for other models

[RYQ-22]

🥰 Feature Description

Currently, only the deepseek-reasoner model supports streamWithThink. However, when we use a third-party DeepSeek API through the OpenAI interface, streamWithThink is disabled.

🧐 Proposed Solution

I have a naive idea but I'm not sure if it‘ll work: detect <think> in stream for all models and turn on "think" mode automatically.

📝 Additional Information

No response

[ninirobot]

能否也支持一下谷歌模型的思考过程？谢谢！

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

! [Image] (https://github.com/user-attachments/aSSETS/E037AB30-5F53-410d-b74b-a537ca019612))
Can you also support the thinking process of Google model? Thanks!

[jackalcooper]

related: #6141

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

related: #6141

[jackalcooper]

在 #6186 修复了格式

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

Fixed format at #6186

[xzcxzcyy]

在 #6186 修复了格式

可以做成一个通用的方案吗？现在看起来只对SiliconFlow生效。

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

Fixed format in #6186

Can it be made into a general solution? Now it looks like it only works for SiliconFlow.

[jackalcooper]

在 #6186 修复了格式

可以做成一个通用的方案吗？现在看起来只对SiliconFlow生效。

因为只有openai原版api带有的的规范变成通用的自定义endpoint才比较合适，现在这个field是deepseek 特有的，所以不太好一口气全改（比如有些模型输出有可能是<think>）。现在的代码本身是通用的，每个模型实现自己的parseSSE函数。

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically.

---

Fixed format in #6186

Can it be made into a general solution? Now it looks like it only works for SiliconFlow.

Because it is more appropriate to only the specifications included in the original openai API become general custom endpoints, this field is now unique to deepseek, so it is not easy to modify them all in one go (for example, some model output may be ). The code now is general in itself, and each model implements its own parseSSE function.