stop the worker thread and track the status, ignoring io requests (i.e.
prevent red_dispatcher writes to red_worker) if red_worker is stopped.

This fixes spice red_worker asserts from actions attepted after X
has destroyed a screen. X runs in a loop destroying and recreating screens
multiple times during the process lifetime, whenever the last X client
has exited.

Note that the spice server is kept up during a X screen restart. This is
on purpose, to avoid a spice-client closing when doing a window manager
restart. Cleanup of the spice-server on atexit - still to do.