Allow multiple usb cameras to be processed in parallel

[rodja]

This PR reworks the usb camera internals to allow seamless usage of multiple v4l2 devices. There had to be made quite a lot of changes to make this possible:

• improving layout and usability of the camera control example (which was used for testing)
• make sure we only add real capturing devices (linux often creates two devices (video0 and video1 for a single usb camera, where only video0 can be used to capture data)
• set mjpeg format for usb cameras before trying to connect -- otherwise the bandwidth might not be sufficient for multiple cameras to get connected
• use linuxpy.video.device instead of opencv -- because opencv auto-detection for devices does not work in all cases and often comes without gstreamer support in containers; also linuxpy provides an async API while cv2 does not
• simplified UsbDevice and renaming it to V4l2Device to make clear that it uses the v4l2 APIs from linuxpy.video.device

ToDos

• Is the frame rate fast enough?
• Implement/rethink setting of width/height/fps/exposure/auto-exposure for V4l2Device.
• Test with four cameras.
• Test/implement/verify YUV fallback with a single camera.

NOTE: if cameras only show black images they might have gotten into an undefined state. Try unplug/replug them or restart the host system.