fix(llm) : Streaming output generation and update gradio to latest version

[Aryankb]

Fixes #173
Fixes:-

• Enabled streaming output generation in rag_block -> rag_answer, can be applied to all other output boxes of gradio
• updated gradio to 5.20.0, no breaking changes found.
• exploring more gradio features, will try to update them with future issues.

Modifications:-
used gr.update() in rag_block to generate streaming output.

How to Test:-
Run python3 -m hugegraph_llm.demo.rag_demo.app in CLI
click Answer question under RAG and user functions.
the output is streaming. (though it is not async. the final output is streaming after collecting all outputs)
Screencast from 03-07-2025 01:13:46 PM.webm

[Aryankb]

@imbajin
i tried my best to have async streaming. but due to many different blocking functions, not able to do it.

[imbajin]

i tried my best to have async streaming. but due to many different blocking functions, not able to do it.

@Aryankb Thank you for your attempt, we already implement a partially less elegant asynchronous streaming output by modifying the previous version.

We can keep this implementation first and let the current PR only upgrade the gradio version. If possible, a simple list of the useful features of 4.x -> 5.x for us is better.

[Aryankb]

@imbajin okay, so i just need to list the possible changes (useful features) or need to make changes in gradio?

Also,
After which PR merge, async streaming was working? Because I fetched the main branch yesterday, it was not working, so i made these changes.
And now it is streaming but not async (after my changes)

[imbajin]

@imbajin okay, so i just need to list the possible changes (useful features) or need to make changes in gradio?

Also, After which PR merge, async streaming was working? Because I fetched the main branch yesterday, it was not working, so i made these changes. And now it is streaming but not async (after my changes)

refer #190 (Added async func calls to the entire path, but some places have TODO)

After introducing the async workflow framework, we should make all current operator/step calls async. Currently, there are many synchronous places, so it is very troublesome to change them