KeyHandleMatcher might theoretically not handle re-used dev_refs correctly

[ttaubert]

From KeyHandleMatcher.update():

// TODO In theory, it might be possible to replace a token between
// `update()` calls while reusing the device_ref/path. We should refactor
// this part of the code and probably merge the KeyHandleMatcher into the
// DeviceMap and Monitor somehow so that we query key handle/token
// assignments right when we start tracking a new device.

I'd certainly favor coming up with a nicer structure for the KHM, DevMap, and Monitor that fits the runloop we have better, yet still maintains some separation of concern.

[ttaubert]

I'd probably favor a solution like in #52, that would allow us to get rid of the KHM completely. Haven't thought about how easy that is with either Windows/Linux. macOS doesn't use the KHM anymore.