Fix test_permutations() in P0202R3_constexpr_algorithm_and_exchange

[StephanTLavavej]

First, this simplifies the P0202R3_constexpr_algorithm_and_exchange test by avoiding a huge unnecessary && chain. When performing "dual mode" testing at runtime and compiletime, only the top-level function needs to be constexpr bool returning true (even that isn't strictly necessary, but it's convenient). The helper functions can be constexpr void; any failures will still be detected by failed asserts.

Second, this adds a missing call to test_permutations().

Third, this fixes test_permutations() which never worked.

• Regardless of whether DevCom-892153 is a compiler bug, we don't need <array>; we can use a built-in multidimensional array with a size_t cursor.
• int buff[] = {1, 2, 3, 4}; needs to be int buff[] = {10, 20, 30, 40};.
• This says constexpr int expected[24][4] instead of constexpr int expected[][4] for clarity. I didn't bother to extract 24 as a magic number, although I could do that if anyone wants.
• After cycling through next_permutation(), this verifies that buff has wrapped around to be equal to expected[0].
• Previously, after assert(!prev_permutation(begin(buff), end(buff)));, this dereferenced cursor == end(expected). Now we reset cursor = 23;.
• After cycling through prev_permutation(), this attempted to verify assert(is_sorted(begin(buff), end(buff)));. That's incorrect - the array is in ascending order after cycling through next_permutation(), but in descending order after cycling through prev_permutation(). I have replaced this with stricter tests: we inspect both cursor and verify that we have exactly expected[23]'s content (this is stricter than verifying is_sorted() with greater<>{}).
  • Technically, this final check of the array's contents is redundant (we performed it during the first iteration of the do/while loop after assert(!prev_permutation(begin(buff), end(buff)));) but it seems reasonable to maintain consistency with the next_permutation() code above.

[AdamBucior]

LGTM (although I already fixed test_permutations in #2043).

[StephanTLavavej]

@AdamBucior Ah, my apologies for partially duplicating your work! I noticed that this test was incorrect last night (it came to my attention because it was suppressing clang-format) and didn't realize that your PR had already fixed it.

I believe that our changes can be combined - I'd like to proceed with this PR's more extensive overhaul, and then #2043 can add the is_permutation() checks (both for (const auto& arr : expected) near the top, and the specific examples near the bottom). They'll need manual merge conflict resolution but I believe they can be completed in either order.

[CaseyCarter]

<meme>I don't always fail to run broken STL test cases, but when I do ...</meme> (No change requested.)

[StephanTLavavej]

I'm mirroring this to an MSVC-internal PR. Please notify me if any further changes are pushed.