x/crypto/argon2: Add AVX2 implementation

[aead]

This is a feature request for adding an AVX2 implementation of Argon2. Currently there is a generic / pure Go and an amd64 SSE4.1 implementation.
An AVX2 implementation can improve performance about ~30 %. The AVX2 implementation requires Go 1.10 because of some instructions which are not supported by the assembler in earlier releases.

See also: google/fscrypt#82

/cc @FiloSottile

[gopherbot]

Change https://golang.org/cl/102422 mentions this issue: argon2: add AVX2 implementation