recalc_window_features() and has_*_func and 
meta_window_maximize etc. needs reworking.

Right now we call recalc_window_features anytime 
we change something, and meta_window_maximize
doesn't check e.g. whether the window is not maximizable
due to its type (the caller is supposed to do it).

I think recalc_window_features should be changed to a 
queue_recalc sort of setup, and we should impose invariants on 
allowed combinations of features (e.g. you can't have a window 
that is both fullscreen and shaded).

There are also cases such as maximized-and-shaded is allowed, but 
maximizing a window has the side effect of unshading it (if the user 
does the maximization, but not when restoring from a session).

We may want a meta_window_user_maximize() and meta_window_real_maximize()
kind of setup here as well.