This uses the stdlib's method of creating a tempfile in our write
file atomimc method, with a few modifications. We use a 64 bit number
rather than 32 bit, and therefore a corresponding LCG. This is to
reduce collision probability. (Note we currently used 32 bytes previously,
so this is likely a concern)
We handle reseeding the LCG in such a way that multiple threads are
even less likely to reuse the same seed.