Improved ripemd160 performance

* cipher/rmd160.c (transform): Interleave the left and right lane
rounds to introduce more instruction level parallelism.

The benchmarks on different systems:

Intel(R) Atom(TM) CPU N570 @ 1.66GHz
before:
Hash:

|  nanosecs/byte   mebibytes/sec   cycles/byte

RIPEMD160 | 13.07 ns/B 72.97 MiB/s - c/B
after:
Hash:

|  nanosecs/byte   mebibytes/sec   cycles/byte

RIPEMD160 | 11.37 ns/B 83.84 MiB/s - c/B

Intel(R) Core(TM) i5-4670 CPU @ 3.40GHz
before:
Hash:

|  nanosecs/byte   mebibytes/sec   cycles/byte

RIPEMD160 | 3.31 ns/B 288.0 MiB/s - c/B
after:
Hash:

|  nanosecs/byte   mebibytes/sec   cycles/byte

RIPEMD160 | 2.08 ns/B 458.5 MiB/s - c/B

• Signed-off-by: Andrei Scherer <andsch@inbox.com>