The generally recommended way is to include it from source files, not
headers. We had a mix of both. Let's try to be consistent and follow the
recommended way.

macOS actually uses the alpha channel on windows, so we can get visual
artifacts if we feed it bogus alpha data. This filtering unfortunately
causes some CPU usage, but it's necessary until we can make sure the
framebuffer always contains proper 0xff for alpha.

We won't always be on the primary monitor, so check which color space
we're actually using right now. For offscreen stuff we assume a standard
sRGB color space.

The system expects these to be immutable, so changing the data after
creation only works in some special cases. We need to recreate the
CGImage object each time we've changed something.

It seems to be more efficient.

Some platforms draw directly to the screen, which means that updates
will flicker if we draw multiple layers. Prevent this by first
composing the update on a hidden surface.

This will allow us to render more things than just the framebuffer.