The button state now indicates if the next video will be started in
fullscreen, or there is already one video in fullscreen.

This removes unused pointering. Furthermore, for nonembedded windows
the video size is set directly to avoid strange resizing after window
creation.

This was not a particularly good idea in user space.

The code is complete and utter crap either way.

(Single precision is more than enough in this case.)

The stream is always killed at the same time as the access already.