Upgrade the random.sql regression test.

Upgrade the random.sql regression test.

We had some pretty ad-hoc and inefficient code here.  To make
matters worse, it didn't test the properties of the random()
function very thoroughly, and it had a test failure rate of
one in every few tens of thousands of runs.  Replace the
script altogether with new test cases that prove much more
about random()'s output, run faster, and can be calculated
to have test failure rates on the order of 1e-9.

Having done that, the failure rate of this script should be
negligible in comparison to other causes of test failures,
so remove the "ignore" marker for it in parallel_schedule.
(If it does fail, we'd like to know about that, so "ignore"
was always pretty counterproductive.)

Tom Lane and Dean Rasheed

Discussion: https://postgr.es/m/4173840.1673290336@sss.pgh.pa.us