Use a polling thread per device

[almarklein]

This PR makes each device create a thread for polling the native device to resolve the promises. This means there are no busy-loops; i.e. its less strain on the CPU while obtaining faster responses.

The implemented technique is heavily inspired by the implementation in the Servo browser (servo/servo#32266). The Firefox browser will likely adopt a similar strategy soon (https://bugzilla.mozilla.org/show_bug.cgi?id=1870699).

When passing a loop instance, this work relies on loop.call_soon_threadsafe from pygfx/rendercanvas#146.

[almarklein]

I don't think anyone uses feature to set the loop yet, but just in case, let's merge this after pygfx/rendercanvas#146 is merged.

[almarklein]

I made a small improvement to the _sync_wait() logic.

I can already see that the FPS in pygfx/rendercanvas#138 increases a bit because the system simply reacts immediately (instead of the previous loop that repeatedly sleeps and checks status). Also, while FPS is nice, the greater advantage IMO is minimizing the delay between rendering a frame and putting it under the user's eyes. This delay has a big impact on the perceived responsiveness.