[x86] Implement v16i16 support with AVX2 in the new vector shuffle