Experiment with shared pages

[falkoschindler]

Motivation

Inspired by discussion #5201, I asked myself (and Sonnet 4.5) what it takes to share NiceGUI pages.

Implementation

A core idea is that we don't need to immediately evaluate the page function, but we can wait until it is visited for the first time. Instead of re-creating a new client on every visit, we store it and re-use it for subsequent visits. To avoid cyclic references (bad for reference counting), we use a weak reference. The resulting implementation is rather simple.

At the moment this is only an experiment.

Example usage:

@ui.page('/', shared=True)
def page():
    ui.label(f'The time is {time.time()}')
    ui.input('Enter your name').on('keydown.enter', lambda e: ui.label(f'Hello, {e.sender.value}!'))
    ui.button('Notify', on_click=lambda: ui.notify('Hello, world!'))

Progress

• I chose a meaningful title that completes the sentence: "If applied, this PR will..."
• The implementation is complete.
• Pytests have been added (or are not necessary).
• Documentation has been added (or is not necessary).

[rodja]

Interesting! But it will re-introduce some of the strangeness which we encountered with auto-index client:

1. all visitors have the request object of the first visitor (and therefore also the same client.ip)
2. client cleanup (eg. pruning) requires special attention
3. what about ui.download?
4. will app.storage.tab work?
5. how to handle returns of await ui.run_javascript? Will there be multiple return values?

[evnchn]

By principle, one-to-one actions will not work when the Python<->Browser relationship is one-to-many.

I foresee we need to state very clearly that when shared=True stuff may break and that is just an inherent limitation, and possibly for the user's sake we may need to add back the checks we dropped in #5005.

Nevertheless I think this PR is good. Second-class support is better than no support, I guess. I'll never look back for my code, though.