Use the compute intermediate message digest (KIMD) instruction
when possible. Adds test to check fallback code path in case
KIMD is not available.

Benchmark changes:
Hash8Bytes  3.4x
Hash1K      9.3x
Hash8K      10.9x

Change-Id: Ibcd71a886dfd7b3822042235b4f4eaa7a148036b
Reviewed-on: https://go-review.googlesource.com/22350
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>