On amd64 we optimize  encoding/binary.BigEndian.PutUint{16,32,64}
into bswap + single store, but strangely enough not LittleEndian.PutUint{16,32}.
We have similar rules, but they use 64-bit shifts everywhere,
and fail for 16/32-bit case. Add rules that matchLittleEndian.PutUint,
and relevant tests. Performance results:

LittleEndianPutUint16-6    1.43ns ± 0%    1.07ns ± 0%   -25.17%  (p=0.000 n=9+9)
LittleEndianPutUint32-6    2.14ns ± 0%    0.94ns ± 0%   -56.07%  (p=0.019 n=6+8)

LittleEndianPutUint16-6  1.40GB/s ± 0%  1.87GB/s ± 0%   +33.24%  (p=0.000 n=9+9)
LittleEndianPutUint32-6  1.87GB/s ± 0%  4.26GB/s ± 0%  +128.54%  (p=0.000 n=8+8)

Discovered, while looking at ethereum_ethash from community benchmarks

Change-Id: Id86d5443687ecddd2803edf3203dbdd1246f61fe
Reviewed-on: https://go-review.googlesource.com/95475
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>