Rearrange cmac's block_mulx, make it closer to xts_shift.

* xts.c (xts_shift): Arrange with a single write to u64[1].
* cmac.c (block_mulx): Rewrite to work in the same way as
xts_shift, with 64-bit operations. XTS and CMAC use opposite
endianness, but otherwise, these two functions are identical.