orient: refactor for SIMD

There are several different approaches to implement transforms in SIMD:

• reverse (h-flip) and transpose instructions, e.g., with ARM NEON,
• register-stride load/store, e.g., with ARM SME or RISC-V V,
• table lookups.

In any case, SIMD is only needed for h-flip and transposition. The other ones can be reconstituted:

• V-flip can be achieved with memcpy.
• 180 degrees rotation is h-flip with one negated stride.
• 90 or 270 degrees rotations is transposition with one negated stride.
• Anti-transposition is transposition with both strides negated.

This MR cuts the number of plane transform callbacks down from 21 to just 6 to simplify future SIMD work. It then also adds two callbacks for 64-bit transforms.

This is a dependency of !1430 (merged)