Reimplement rows case in zip_mut_with using Zip

[bluss]

• Reimplement the zip by rows case using Zip.
• Improves performance of the most important non-contiguous array shape: Where rows are contiguous but we have a greater stride from row to row. This appears if you split a contiguous array into rectangular pieces.

[bluss]

Ups and downs:

 name zipbefore.0 ns/iter zipafter.0 ns/iter diff ns/iter diff % add_1d_regular 450 447 -3 -0.67% add_1d_strided 3,215 3,293 78 2.43% add_2d_0_to_2_iadd_scalar 353 355 2 0.57% add_2d_alloc 979 904 -75 -7.66% add_2d_assign_ops 467 445 -22 -4.71% add_2d_broadcast_0_to_2 357 353 -4 -1.12% add_2d_broadcast_1_to_2 641 542 -99 -15.44% add_2d_cutout 916 783 -133 -14.52% add_2d_cutouts_by_16 2,189 1,626 -563 -25.72% add_2d_cutouts_by_32 1,303 1,046 -257 -19.72% add_2d_cutouts_by_4 9,881 6,814 -3,067 -31.04% add_2d_f32_regular 452 451 -1 -0.22% add_2d_regular 452 443 -9 -1.99% add_2d_strided 2,155 2,593 438 20.32% add_2d_strided_dyn 6,696 3,058 -3,638 -54.33% add_2d_transposed 2,094 2,585 491 23.45% add_2d_zip 444 461 17 3.83% add_2d_zip_alloc 803 840 37 4.61% add_2d_zip_cutout 902 866 -36 -3.99% add_2d_zip_strided 2,563 2,578 15 0.59% add_2d_zip_transposed 2,549 2,587 38 1.49% assign_scalar_2d_corder 651 649 -2 -0.31% assign_scalar_2d_cutout 857 876 19 2.22% assign_scalar_2d_forder 653 650 -3 -0.46% assign_zero_2d_corder 608 626 18 2.96% assign_zero_2d_cutout 895 875 -20 -2.23% assign_zero_2d_forder 625 625 0 0.00% iadd_2d_regular 353 355 2 0.57% iadd_2d_strided 1,568 1,600 32 2.04% scalar_add_1 745 741 -4 -0.54% scalar_add_2 737 743 6 0.81% scaled_add_2d_f32_regular 533 539 6 1.13%