optimize large elements for slice rotating

[ywxt]

This PR introduced two fixes to improve performance of slice::rotate.

Fix A:

For large elements, interrupter them as *MaybeUninit<usize>, which makes them choose the swap algorithm and obtain better cache/TLB locality.

Fix B:

When a side in swap algorithm is tiny, it might be a lot of small swaps. The fix B switches tail swap to memmove.

Benchmark	Original (ns)	Fix A (Δ)	Fix B (Δ)	Fix A+B (ns)	A+B Improvement
rotate_medium_half_plus_one	6267.62	1779.07 (+71.6%)	6323.59 (-0.9%)	1621.24	+74.1%
rotate_huge_by9199_big	8218230	8109777 (+1.3%)	4281360 (+47.9%)	4244542	+48.4%
rotate_huge_half_plus_one	6268406	4799403 (+23.4%)	6221231 (+0.8%)	4758676	+24.1%
rotate_huge_by1234577_big	7950044	7846722 (+1.3%)	6111909 (+23.1%)	6102005	+23.2%
rotate_medium_by727_u64	3702.35	3606.24 (+2.6%)	3139.26 (+15.2%)	3031.55	+18.1%
rotate_medium_by727_bytes	3704.06	3605.89 (+2.7%)	3147.87 (+15.0%)	3033.77	+18.1%
rotate_medium_by727_strings	3608.34	3578.12 (+0.8%)	3035.60 (+15.9%)	3005.55	+16.7%
rotate_medium_half	1474.78	1476.64 (-0.1%)	1370.02 (+7.1%)	1354.31	+8.2%
rotate_huge_half	3179321	3150365 (+0.9%)	3077309 (+3.2%)	3038567	+4.4%
rotate_huge_by9199_bytes	4636449	4568776 (+1.5%)	4509977 (+2.7%)	4477535	+3.4%
rotate_huge_by1234577_strings	6276064	6182101 (+1.5%)	6090534 (+3.0%)	6076560	+3.2%
rotate_huge_by1234577_bytes	6266812	6218410 (+0.8%)	6236837 (+0.5%)	6097584	+2.7%
rotate_huge_by9199_strings	4727975	4698212 (+0.6%)	4676240 (+1.1%)	4610863	+2.5%
rotate_huge_by1234577_u64	6318677	6264397 (+0.9%)	6236986 (+1.3%)	6169595	+2.4%
rotate_huge_by9199_u64	4594580	4516849 (+1.7%)	4510532 (+1.8%)	4527773	+1.5%
rotate_medium_by1	1486.18	1469.57 (+1.1%)	1468.76 (+1.2%)	1467.72	+1.2%
rotate_huge_by1	3327706	3327696 (~0%)	3319754 (+0.2%)	3290266	+1.1%
rotate_tiny_by1	8.28	8.64 (-4.3%)	8.64 (-4.3%)	8.63	-4.2%
rotate_tiny_half	7.94	8.29 (-4.4%)	8.29 (-4.4%)	8.28	-4.3%
rotate_tiny_half_plus_one	9.41	9.41 (0%)	9.43 (-0.2%)	9.69	-3.0%

r? @Amanieu

[rustbot]

⚠️ #[rustc_allow_const_fn_unstable] needs careful audit to avoid accidentally exposing unstable
implementation details on stable.

cc @rust-lang/wg-const-eval