Refactor resizing logic and avoid mismatch between last resize event and draw

[almarklein]

Fixes #56

Instead of backends needing to implement _rc_get_physical_size, _rc_get_logical_size and _rc_get_pixel_ratio, they must now call _set_size_info() whenever their size changes. Backends should also not submit a resize event, the base canvas now handles that.

The base class will now emit a resize event right before processing other events, but also right before doing the draw, if the size has changed since events were processed. This fixes #56.

Another benefit is that the backends become simpler, and the getters like get_physical_size() may be faster because values are read from the base class's cache, instead of calculated from the backend.

[kushalkolar]

@FlynnOConnell is excited to stop whatever he's doing and try this right now 😂

[FlynnOConnell]

🙇‍♂️ thank you!

[FlynnOConnell]

Confirmed no more resizing errors! Unless you reduce the size to nothing, which is normally not an issue

rendercanvas.mp4

[kushalkolar]

Unless you reduce the size to nothing, which is normally not an issue

Ah this is because the canvas has an actual size but portion of the canvas pygfx draws to has become negative. I wonder if there's a min window dims we can set.

[almarklein]

Unless you reduce the size to nothing

In this case it's FPL that does not apply bounds to the viewport :)

But a zero size to the canvas can in theory also happen. I added a note regarding that.

[almarklein]

@Vipitis I'll take care of adjusting #115