mpz_fac_ui (revised)

[Paul Zimmermann]

   The sieving tables might be possible to shrink to half their
   current size if only odd numbers were represented.  Or would that
   cost too much overhead?

The optimal would be to shrink them so that they do not exceed the output
size (at least for large n), and sieve in k subintervals of length n/k.
This is possible, but will complicate the code.

   Other issues:

   1. Could (n over k) be computed similarly?  Want to work on that?

I guess the same idea should work for (n over k).

   2. Some more comments wouldn't hurt.

Right. Today's version already had much more comments.

   3. How about speed for small operands, does it beat the current,
      complex code already from 3!  If not, we might want to have a
      base case under a threshold.

Right. Of course the optimizations in the current code apply also here.

Paul