I talk a little about why in my "fearless SIMD" blog post (also introducing a prototype that wraps the unsafety but at the moment is very limited in exactly what SIMD operations are exposed). https://raphlinus.github.io/rust/simd/2018/10/19/fearless-si...
I don't see why they used unsafe in `add_assign` - an assert!(octets.len() == other.len()) would likely have elided the bounds checks.