Implement ranges::sample and ranges::shuffle

[miscco]

This ports the sampling algorithm to ranges.

I am unsure what exactly the problem is with the tests. I guess somecombinations of the machinery do not play well with the requirements but I have to figure out which.

[StephanTLavavej]

Thanks for the initial review, @statementreply! 😸 Moving to WIP as you've identified the cause of (at least one of) the test failures (mentioning a nonexistent _It).

[miscco]

It seems activating the tests actually helps...

[miscco]

Given that we move the concept definition into algorithm I have put the ranges::shuffle implementation into this PR as otherwise merging is a hassle and both are similar anyway

[miscco]

Tests are active and passing so this should be ready for review

[CaseyCarter]

Just a nitpick which I'll go ahead and apply.

[StephanTLavavej]

Looks good modulo comments, thanks!

[CaseyCarter]

Thanks for implementing these algorithms, whose choice I suspect was actually quite deliberate - not random at all.

[eugnsp]

The new loop termination condition in Sample_selection_unchecked is wrong. The loop should be terminated when Count hits zero, not PopSize. The current implementation leads to redundant RngFunc(PopSize) calls, just to check the condition RngFunc(PopSize) < 0, which will never be satisfied. This results in redundant calls to a random number generator after the sampling is already done and to unreasonable deviations in behaviour between 16.7 and 16.8. The average number of such redundant calls is about [InputRangeSize] / ([OutputRangeSize] + 1), which could be significant if a small number of elements is requested.