<xutility>: optimize _Find_unchecked for data sizes other than 1

[AlexGuteniev]

_Find_unchecked is the implementation of std::find.

Currently it is optimized for 8 bit size array elements by using memchr.

It is possible to expand this optimization for 16 bit size array elements by using wmemchr.

For greater size operands, like 32 or 64 bit, it is possible to implement the optimization either, but it takes some manual implementation.

Assuming 32 bit element, the comparison using SSE2 can be made with _mm_cmpeq_epi32. Then with _mm_movemask_epi8, mask can be extracted. If it is nonzero, countr_zero may be used to determine the position of the first match. This only requires SSE2, which is x64 baseline.

With AVX2, there are 256-bit variables available and 64 bit data sizes are possible.

The implementations should probably go to vector_algorithm.cpp

[AlexGuteniev]

Note there's a PR currently in review in the same area: #2380

[AlexGuteniev]

Correction, we can't rely on wmemchr as it is not vectorized and thus slow.
This is reported as DevCom-1614562.
Need to wait till that is fixed, or we can implement 16-bit version on our own as well.

[AlexGuteniev]

Inspected memchr under the debugger.
Apparently it implements exactly the algorithm I suggest.
But it is just SSE2, and we deserve AVX2 already.
Besides, it suffers from DevCom-1615707, which is relevant for ranges find with unreachable_sentinel, and may be relevant elsewhere.

So, looks like should implement a custom solution even for size 1.

[AlexGuteniev]

Relevant test: std\tests\Dev11_0316853_find_memchr_optimization
Also ranges find test: std\tests\P0896R4_ranges_alg_find
Also parallel find test: std\tests\P0024R2_parallel_algorithms_find