Handle zero-sized buffers internally

[madsmtm]

Use u32 in our public API instead of NonZeroU32, and handle width == 0 || height == 0 internally. This resolves #238. Note that there are still a few inconsistencies between platforms in how things are rendered when the buffer is smaller than the window, but that is pre-existing, see #177.

We can now also default to a 0-sized buffer instead of panicking in buffer_mut, which resolves #106.

Tested on:

• Android
• CoreGraphics
• KMS/DRM
• Orbital
• Wayland
• Web
• Win32
• X11

[ids1024]

We can attach a null buffer to a surface on Wayland, but this has the effect of unmapping the surface, rather than having a surface with a 0x0 size. If that isn't how this behaves on other platforms, it may be better to use a 1x1 (transparent?) buffer to approximate the effect of having the surface mapped but with a 0x0 size.

[madsmtm]

Yeah, I've been kinda going back and forth between what would be best here.

On macOS, if you don't provide any CALayer contents, whatever is behind the surface is rendered (usually the window backdrop). If the window is marked as transparent (the window backdrop is made transparent), only the title bar will remain.

IIUC that's basically what using a 1x1 buffer would do on Wayland, whereas unmapping the surface also means that the window decorations etc. are removed.