mwm: Handle as many events as possible before redrawing

mwm: Handle as many events as possible before redrawing

The current implementation redraws the monitors (where necessary) after
each handled event. This leads to problems when an application destroys
or unmaps multiple windows in one go, because mwm will attempt to redraw
the destroyed windows for which it hasn't processed the destroy/unmap
event yet.
This commit changes the mwm_run() function to handle all queued events
before redrawing anything.