New random code

[Torbjorn Granlund]

I have finally implemented the mpz_rrandomb code I suggested when
working with Niels on testing his new GCD code.

The old code would make rows of just a few dozens of bits, which
might be adequate for triggering ripple phenomena in operations
on numbers of up to a few thousand bits.

The new mpz_rrandomb makes the row length (randomly) propotional
to the size of the generated numbers.

This new mpz_rrandomb immedately triggered a bug in the FFT
multiply code.  This was no big surprise, given the bug report
from earlier this week.  It is very likely the same bug.

In order to avoid havoc with the daily builds, I will temporarily
disable the FFT code in the daily builds.

(It will be interesting to see if the new code can trigger other
bugs.  We have been almost spared from miscomputation bugs in
released code.)

-- 
Torbjörn