poly1305: SSE2 and AVX2 acceleration

[tarcieri]

The poly1305 crate presently only provides a portable software implementation.

Ideally we'd also provide SSE2 and AVX2 implementations which can be translated from an existing robust C/C++ implementation based on Intel intrinsics.

The best there is probably @floodyberry's https://github.com/floodyberry/poly1305-opt

• SSE2: https://github.com/floodyberry/poly1305-opt/blob/master/sources/poly1305-donna-x64-sse2-incremental-source.c
• AVX2: https://github.com/floodyberry/poly1305-opt/blob/master/sources/poly1305-donna-x64-avx2-incremental-source.c

[str4d]

An AVX2 backend has been merged in #49. It isn't based on poly1305-opt, so one possible next step is to refactor that AVX2 code along the lines of the new backend's types, and see how similar (or not) it is.

[tarcieri]

I think we can close this now that #49 is merged and poly1305 v0.6.1 is released.