Streamable pages with yield

[evnchn]

Motivation

The gist of pipeline parallelization is that, should a task involve several steps which could have been in parallel but must be completed serially for each task, we want to break them down so that the parallelization takes place.

• There are multiple steps for rendering a NiceGUI page: Page function execution, network transmission
• But the page function is monolithic and the respinse is also monolithic 😞

This PR challenges the status quo.

Implementation

Even if nothing is done, there is already speedup, because the template is broken into 2, and now we have 3 parts:

• Before elements string
• elements string
• After elements string

On top, this code is now possible:

@ui.page('/')
async def main():
    for _1 in range(5):
        for _2 in range(10000):
            ui.label('1')
        await asyncio.sleep(0.1)
        yield

Here, we have 5 chunks of 10000 elements, each generated 0.1 second apart.

Fast 4G

Before this PR: 2.38s
After this PR: 1.73s

Progress

• I chose a meaningful title that completes the sentence: "If applied, this PR will..."
• The implementation is complete. (I'm not sure)
• Pytests are very necessary and pending.
• Documentation is necessary and pending.
• While the math checks out, we should do more testing to see if the time savings are worth it in real life. (2.38 vs 1.73)
• Do we need explicit stream=True or can we infer?
• How does this compare to just await client.connected() then push content changes through the outbox?
  • Push via outbox will lead to rendering and is slow that way.