Seeding of GMP test suite

[Torbjörn Granlund]

Each night, we run each GMP test with almost 1000 seeds.  The seeds are
32 bits.

If we assume the seeds are perfectly uniformly distributed, we can
expect to start having collisions after about 80 days.  Since we've been
running this for many years (although not as intensively as now) we
should be getting a good number of collisions.

Some collisions are no problem, of course.  But they will gradually make
are testing less effective.

We should probably add a handful more bits to the seeds, and make sure
all seed bits are actually used.

-- 
Torbjörn
Please encrypt, key id 0xC8601622