x86 version using inline assembly; at least gcc-4.4.5 recognizes
shift-and-or expressions which are in fact rotations.
(_nettle_ripemd160_compress): Use LE_READ_UINT32.

Rev: nettle/ripemd160-compress.c:1.2