rand() requires we call srand() instead,
but might as well use random().

As-is, all threads will have the same seed and
(in some cases) quickly encounter problems when
they generate the same temporary link names in the same order.

Looking at the glibc source, it appears to implement
rand() using random() which is why this didn't cause
bigger problems previously.

Note that "random()" does return "long int" instead
of the "int" returned by "rand()" but that seems okay:
either they're the same size or the larger type
results in slightly longer names for our temporary
links bringing with it a corresponding reduction
in chance of collision between threads.