Use memcpy in uninitialized_meow

[AlexGuteniev]

Resolves #6029

memcpy is the same as memmove when it is usual library call.

But intrinsic optimizations are possible for known size for memcpy but not for memmove.
These optimizations happen when size is known at compile time (compile time constant or propagated constant), and this size is small enough (the boundary is at about kilobyte order of magnitude).

Uninitialized algorithms require non-overlapped range: See [uninitialized.copy]/1, and the others. It would have been strange to have overlap between initialized memory and uninitialized memory.

There hasn't been _Copy_memcpy / _Copy_memcpy_n, so I've added them. They might be useful in other places, besides uninitialized algorithms, as some other algorithms have useful non-overlap guarantee, but for now lets focus on uninitialized algorithms.

Benchmark results show overall improvement, although for some larger amount of data that is 800 bytes it flips. Probably the compiler should have its internal threshold to always use the library memcpy set to a lower value. For very large amounts where the compiler does emit memcpy the results before and after the change do match.

Benchmark	Before	After	Speedup
bm_uninitialized_copy<1, uint8_t, highly_aligned>	1.18 ns	1.16 ns	1.02
bm_uninitialized_copy<5, uint8_t, highly_aligned>	1.41 ns	1.17 ns	1.21
bm_uninitialized_copy<15, uint8_t, highly_aligned>	1.63 ns	1.42 ns	1.15
bm_uninitialized_copy<26, uint8_t, highly_aligned>	1.41 ns	1.19 ns	1.18
bm_uninitialized_copy<32, uint8_t, highly_aligned>	1.41 ns	1.18 ns	1.19
bm_uninitialized_copy<38, uint8_t, highly_aligned>	2.32 ns	1.40 ns	1.66
bm_uninitialized_copy<60, uint8_t, highly_aligned>	2.39 ns	1.43 ns	1.67
bm_uninitialized_copy<64, uint8_t, highly_aligned>	2.33 ns	1.41 ns	1.65
bm_uninitialized_copy<125, uint8_t, highly_aligned>	2.33 ns	1.91 ns	1.22
bm_uninitialized_copy<800, uint8_t, highly_aligned>	5.68 ns	7.00 ns	0.81
bm_uninitialized_copy<3000, uint8_t, highly_aligned>	15.0 ns	15.1 ns	0.99
bm_uninitialized_copy<9000, uint8_t, highly_aligned>	41.7 ns	41.9 ns	1.00
bm_uninitialized_copy<1, uint8_t, not_highly_aligned>	1.30 ns	1.18 ns	1.10
bm_uninitialized_copy<5, uint8_t, not_highly_aligned>	1.41 ns	1.18 ns	1.19
bm_uninitialized_copy<15, uint8_t, not_highly_aligned>	1.66 ns	1.40 ns	1.19
bm_uninitialized_copy<26, uint8_t, not_highly_aligned>	1.40 ns	1.16 ns	1.21
bm_uninitialized_copy<32, uint8_t, not_highly_aligned>	1.39 ns	1.17 ns	1.19
bm_uninitialized_copy<38, uint8_t, not_highly_aligned>	2.31 ns	1.42 ns	1.63
bm_uninitialized_copy<60, uint8_t, not_highly_aligned>	2.35 ns	1.88 ns	1.25
bm_uninitialized_copy<64, uint8_t, not_highly_aligned>	2.32 ns	1.87 ns	1.24
bm_uninitialized_copy<125, uint8_t, not_highly_aligned>	2.82 ns	2.85 ns	0.99
bm_uninitialized_copy<800, uint8_t, not_highly_aligned>	5.57 ns	12.4 ns	0.45
bm_uninitialized_copy<3000, uint8_t, not_highly_aligned>	15.0 ns	15.3 ns	0.98
bm_uninitialized_copy<9000, uint8_t, not_highly_aligned>	42.4 ns	42.6 ns	1.00