Fix device watch errors and race-conditions

[hpccc53]

Device watching had some problems with both race-condititions and sensitivity to the expectable unpredictable batching and duplicate file events.

The important commits:

• Ignore irrelevant device events
  • Ignore file events for devices that are already grabbed, those would cause EBUSY when trying to regrab a device.
  • Ignore multiple file events for the same device in same batch.
• Only remove a single device, when events can't be fetched.
  • Instead of ungrabbing all devices, when it's detected one is removed. Only that specific device is removed. This fixes the case where two devices are removed, and it's tried to regrab the second removed device. It's better to react on the information that is certain instead of regrabbing all devices.
• Fix race from InputDevice::try_from() to device.grab()
  • There is no way to atomically grab a device. When reading /dev/input folder, anything can happen in the time until device.grab() is called. The only solution is therefore to squash ENODEV errors when grabbing. Note that InputDevice::try_from() already squashes all errors.

Problems that still remain

evdev prints an error message, if the device can't be ungrabbed, even though the device is gone, and ungrab is not needed. I have made a PR that is still open.

The error message is "Failed to ungrab device: No such device (os error 19)"