SSE, AVX-128 & AVX2 vectorized RAID-Z1/2/3 parity computation.

This is just a proof-of-concept to gauge the interest of doing the parity computation in vector registers.
It has received very little testing. Don't use on production systems, obviously :-)
SSE should work with any x86-64 CPUs. AVX-128 should work on AVX-enabled CPU, i.e. Sandy Bridge and later. AVX2 should work on Haswell and later. The exact variant is picked at runtime.