Bit-permuting 16 u32s at once with AVX-512

[mfiguiere]

http://bitmath.blogspot.com/2024/12/bit-permuting-16-u32s-at-once-with-avx.html

[rurban]

9-bit indices for 512 bits? That's why AVX-256 is so much easier and feels way more natural, with 8 bit indices covering the whole space.