native/macos: Re-re-implemented event loop

[birhburh]

• Reverted macos.rs to this implementation: Implemented blocking_event_loop for osx #443 (it's probably better compare changes with this PR than with latest commit to understand what was added additionally)
• Now using own NSView with NSOpenGLContext instead of NSOpenGLView
• For metal backend using redraw instead setNeedsDisplay, because somehow it reduces cpu usage (Cannot find info about enabling vsync like this in MTKView docs)
• Fixed freezing on resize by drawing in draw_rect that called during "live resize"™. I don't like this approach, but it blocks main event loop while resizing, so it will not be some kind of concurency of opengl stuff i think
• Reducing CPU usage when window is occluded
• Added comments to hacky places, there are lots of them imo

Fixes:

• MacOS doesn't redraw while window resize is held but not moving #455
• MacOS CPU usage increased by 8x in v0.4.4 #470

[not-fl3]

Can't say that I understand a problem enough to have an opinion on the high-level approach, but reading through the code nothing strikes me wrong. I did not found excess of hacks, and with proper comments they did not really looked like too much of hacks anyway. Well, I guess it counts as a review.

In other words, great work, I trust your judgement here, @birhburh, lets merge it and will see whats up with bug reports!

[brettchalupa]

I tested out this latest code on master on my Apple Silicon chip on macOS and it segfaults when running cargo run --release --example triangle. Possibly the same as this segfault on @birhburh's fork's prototype branch #470 (comment) but not 100% sure.

Happy to provide some more details or test things out. Let me know if a new issue is desired. Thanks!