[x86] Implement AVX2 support for v32i8 in the new vector shuffle

lowering.

This completes the basic AVX2 feature support, but there are still some
improvements I'd like to do to really get the last mile of performance
here.

llvm-svn: 218440