Refactor winit_runner to make custom runners more usable

[alice-i-cecile]

What problem does this solve or what need does it fill?

App runners control the main loop of Bevy apps.
Unfortunately, winit also needs to control the main loop of the app, and thus integrating Bevy with winit requires the use of a specialized winit_runner.

This code is entirely monolithic, which makes it very challenging to reuse windowing functionality while using a specialized runner.

What solution would you like?

Refactor the contents of winit_runner and winit_runner_with (which frankly just looks like it shouldn't exist at this point).

The code feels like it can be split up into three parts:

1. A default framework for converting winit WindowEvents into Bevy data.
2. Small, useful public functions that handle the conversion of specific window event types.
3. Looping control.

What alternative(s) have you considered?

I've seen users copy-paste the code into their own runner, and make their tweaks there. This is not great, because it means any improvements or fixes we make aren't incorporated upstream.

Additional context

Discussed in the context of #4530, which will require a serious refactor anyways.

[bjorn3]

winit_runner_with is necessary to run on another thread than the main thread as you need to pass in EventLoop::new_any_thread() in that case. It is also necessary to force a specific window api like X11 or Wayland.

[alice-i-cecile]

Useful context, thanks. We should really document that in the comments.

[fowenix]

After a day of trying to get bevy_winit to work with #4531, i keep getting stuck with borrow checker issues because of the usage of app.world everywhere in the Event::WindowEvent match statement