Try to optimize proxy_func

[almarklein]

In benchmarking for pygfx/pygfx#763 it looked like the proxy_func is called a lot, so I figured that of it could be optimized it would be an easy win. It looks like the result is not really measurable, but nevertheless might be worth applying.

• Use a deque instead of a list.
• Move the capturing logic to the ErrorHandler class, so that the logic in the proxy_func becomes simpler. Also cleaner code imo.
• The above also means that the proxy_func closure only captures the error_handler object and no longer uses self.

[almarklein]

BTW: I tried to find other areas for improving performance. One is that our glfw GUI backend consumes a bit more CPU than Qt, because of the custom event-loop to keep the windows alive. This can be refactored to use asyncio to do more of the scheduling. But let's do this when we work on #355 / #383 / update modes.

[almarklein]

BTW, a Cython-based wrapper for wgpu-native might provide obsevable performance improvement :)