Rework the test so that it fails far less often.

A number of independent tests are executed and 5% are expected to fail.
The number of such failures follows a binomial distribution which permits
a statistical test a 0.01% expected failure rate.

There is a command line option to enable the stochastic range checking.
It is off by default.
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/8830)