Feature: Use AI to polish the response of the conversation.

[Fengrui-Liu]

This is a feature that I find very useful, and I am willing to contribute it to the community.

Feature description: If multiple reply nodes (such as "specified reply" and "AI dialogue") are triggered during the execution of a workflow, the reply content is a concatenation of the strings output by multiple nodes, which is not user-friendly in terms of practical usage experience. This is particularly common in scenarios involving multiple application calls.

Therefore, we have added the AI Reply Polishing feature. When the user enables this feature, all content outputted by intermediate nodes in the workflow will be intercepted and recorded until the end of the workflow, at which point AI will be used to polish and finalize the output.

You can refer to the documentation link: (docSite/content/docs/workflow/modules/ai_polish.md)

[c121914yu]

🤖 Generated by deploy action

👀 Visit Preview

[c121914yu]

非常感谢你的pr。但是FastGPT中似乎已经具备了该功能，您可以在高级编排中通过点击【AI对话】的更多设置，来关闭AI的输出，并将其输出内容在流程结束后，连接到指定回复中。
我不确定这是否和您的功能是一致的。

[c121914yu]

非常感谢你的pr。但是FastGPT中似乎已经具备了该功能，你可以在高级编排中通过点击【AI对话】的更多设置，来关闭AI的输出，将其内容放到输出流程结束后，连接到指定回复中。 我不确定这是否和您的功能是一致的。

我又看了一次您的代码，和我之前理解的不太一样。这个功能我会仔细评估一下。

[RandyZ]

我们在用的过程中也发现了这个问题，关闭LLM节点的输出会导致问题无法向下传递，打开节点输出，结果会串在一起。