crypto/md5: simplify generic implementation

crypto/md5: simplify generic implementation

This change uses library functions such as bits.RotateLeft32 to
reduce the amount of code needed in the generic implementation.
Since the code is now shorter I've also removed the option to
generate a non-unrolled version of the code.

I've also tried to remove bounds checks where possible to make
the new version performant, however that is not the primary goal
of this change since most architectures have assembly
implementations already.

Assembly performance:

name                 old speed      new speed      delta
Hash8Bytes           50.3MB/s ± 1%  59.1MB/s ± 0%  +17.63%  (p=0.000 n=9+8)
Hash1K                590MB/s ± 0%   597MB/s ± 0%   +1.25%  (p=0.000 n=9+9)
Hash8K                636MB/s ± 1%   638MB/s ± 1%     ~     (p=0.072 n=10+10)
Hash8BytesUnaligned  50.5MB/s ± 0%  59.1MB/s ± 1%  +17.09%  (p=0.000 n=10+10)
Hash1KUnaligned       589MB/s ± 1%   596MB/s ± 1%   +1.23%  (p=0.000 n=9+10)
Hash8KUnaligned       638MB/s ± 1%   640MB/s ± 0%   +0.35%  (p=0.002 n=10+10)

Pure Go performance:

name                 old speed      new speed      delta
Hash8Bytes           30.3MB/s ± 1%  42.8MB/s ± 0%  +41.20%  (p=0.000 n=9+9)
Hash1K                364MB/s ± 4%   394MB/s ± 1%   +8.27%  (p=0.000 n=10+10)
Hash8K                404MB/s ± 1%   420MB/s ± 0%   +4.17%  (p=0.000 n=10+9)
Hash8BytesUnaligned  30.3MB/s ± 1%  42.8MB/s ± 1%  +40.92%  (p=0.000 n=9+10)
Hash1KUnaligned       368MB/s ± 0%   394MB/s ± 0%   +7.07%  (p=0.000 n=9+9)
Hash8KUnaligned       404MB/s ± 1%   411MB/s ± 3%   +1.91%  (p=0.026 n=9+10)

Change-Id: I9a91fb52ea8d62964d5351bdf121e9fbc9282852
Reviewed-on: https://go-review.googlesource.com/c/137355
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>