The aesenc instruction has high latency.  For hashing large objects,
hash several streams in parallel.

benchmark                         old ns/op     new ns/op     delta
BenchmarkHash5                    7.02          7.45          +6.13%
BenchmarkHash16                   6.53          6.94          +6.28%
BenchmarkHash32                   8.38          8.26          -1.43%
BenchmarkHash64                   12.6          12.0          -4.76%
BenchmarkHash1024                 247           62.9          -74.53%
BenchmarkHash65536                17335         2966          -82.89%
BenchmarkHashInt32Speed           14.7          14.9          +1.36%
BenchmarkHashInt64Speed           14.6          14.9          +2.05%
BenchmarkHashBytesSpeed           35.4          28.6          -19.21%
BenchmarkHashStringSpeed          22.0          20.4          -7.27%
BenchmarkHashStringArraySpeed     65.8          56.3          -14.44%

Change-Id: Ia8ba03063acc64a9066b8ab2d79f2c9aaac1770f
Reviewed-on: https://go-review.googlesource.com/1330Reviewed-by: Russ Cox <rsc@golang.org>