Bits.unaligned() doesn't handle Solaris amd64 correctly

Bits.unaligned() checks only for os.arch=i386 and os.arch=x86. Thus unaligned moves
aren't used with nio on Solaris amd64 which penalizes it performance-wise.