Study impact of native implementations of polyvec_compress + polyvec_decompress

[mkannwischer]

#224 identified polyvec_compress/polyvec_decompress as major bottlenecks that may benefit from optimization (either assembly or nudging the compiler to do the right thing).
It turned out not to be necessary to do that at least with a recent compiler (gcc14), but it would still be good to study the impact once again and decide if we do want to make changes.

Would be good to study this once again for both x86_64 and AArch64.