While testing for Xinerama during setup was done display-wide, each time
XineramaQueryScreens() is called either in start-up or due to an XRR
event, the library re-tests for the Xinerama extension anyway before
moving on; so the initial test is redundant and allows another global to
go away with one other change: always fill in sc->xinerama (and _no),
regardless of the success of malloc in XineramaQueryScreens(), and use
it to see if Xinerama dimensions exist when asked by client and/or menu
code.