pdiff was hiding a rgb24 failure here, as the test was drawing using
black ink on the default black background. Instead, explicitly fill
the surface with white first.
This was triggering an infinite loop (with 24.8 fixed-point) just before
the previous fix.
NOTE: I usually put bug demonstrations just before the fixes, but this
one was quite harsh---not only was cairo looping infinitely, but it
was appending to an array on each iteration---so not kind at all.
