<algorithm>: randomized test coverage for std::search_n and ranges::search_n

[CaseyCarter]

We've been bitten by bugs in the deceptively simple search_n algorithm in the past, and ranges::search_n makes some slightly different implementation choices from std::search_n. It would raise confidence in the correctness of those choices if we had a test that searched random strings of binary digits for sequences of 0s or 1s of randomly-chosen lengths and validated that ranges::search_n and std::search_n produce identical correct results.

(Such a randomized test must ensure reproducibility and could do so by following the model of tests/std/tests/P0067R5_charconv/test.cpp.)

[AlexGuteniev]

With #5352 we have progress towards this one. We have some random coverage, with the following differences:

1. We don't pretty print the failed case
2. We don't have seed as command-line arg
3. We only have coverage for random iterators, forward iterators branch is not covered
4. We have more random strings than just binary 0 and 1

I expect that 4 is fine. Are all other differences still required to be addressed to resolve this issue?

[StephanTLavavej]

As long as we record the seed data, printing the failed case isn't absolutely necessary. We can also live without a convenience mechanism to pass a seed on the command line (since we can always edit the code to pass the seed). However, if the algorithm has substantially different codepaths for RanIt vs. FwdIt, covering both would be a good idea.