* Change the license to a strong Copyleft.
 * Use go modules.
 * Change the canonical upstream to gitlab.
 * Refactor to be more nicely abstracted.
 * `NewCipher` -> `New`.
 * Force bounds-check-elimination in strategic locations.
 * (ref) Assume that `math/bits` always exists.
 * (ref) Don't bother with unsafe anymore, use hardware implementations.
 * (amd64) Don't bother with PeachPy anymore.
 * (amd64) Use Andrew Moon's assembly as the base of implementations.
 * (amd64) Use a vectorized HChaCha.
 * (amd64) Use SSSE3 instead of SSE2 for the non-AVX2 code.