Add AVX512 GFNI processing

[klauspost]

Master/This on Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz:

$go test -test.run=None -bench=Parallel_8x8x05M -cpu=1,2,4,8,16,32,64,128
...
benchmark                         old ns/op     new ns/op     delta
BenchmarkParallel_8x8x05M         1101343       408522        -62.91%
BenchmarkParallel_8x8x05M-2       541933        197688        -63.52%
BenchmarkParallel_8x8x05M-4       278581        93755         -66.35%
BenchmarkParallel_8x8x05M-8       138552        57332         -58.62%
BenchmarkParallel_8x8x05M-16      73836         41346         -44.00%
BenchmarkParallel_8x8x05M-32      69233         35899         -48.15%
BenchmarkParallel_8x8x05M-64      89550         38715         -56.77%
BenchmarkParallel_8x8x05M-128     96317         48993         -49.13%

benchmark                         old MB/s      new MB/s      speedup
BenchmarkParallel_8x8x05M         7616.70       20534.04      2.70x
BenchmarkParallel_8x8x05M-2       15479.04      42433.58      2.74x
BenchmarkParallel_8x8x05M-4       30111.92      89473.49      2.97x
BenchmarkParallel_8x8x05M-8       60544.66      146315.58     2.42x
BenchmarkParallel_8x8x05M-16      113610.76     202887.60     1.79x
BenchmarkParallel_8x8x05M-32      121164.99     233673.15     1.93x
BenchmarkParallel_8x8x05M-64      93675.25      216675.22     2.31x
BenchmarkParallel_8x8x05M-128     87093.32      171221.14     1.97x

AVX512 by itself was actually slower than current AVX2 code, so it was removed.