Two more null randr screen private checks

[Keith Packard]

Ok, checking the call trace sent in by Ricardo Salveti, I figured out
that the problem was an X server with some screens with randr and some
without. I went and audited the randr code and found two more possible
places where an application could get the X server to dereference a
NULL pointer due to this problem.

I also include here a patch which cleans up all of the remaining
compiler warnings in the randr directory.

-keith