Optimized assembly acceleration

[tarcieri]

Though we should definitely maintain a baseline profile which is pure Rust, it would be nice to be able to leverage optimized assembly implementations, and ideally formally verified ones which have been mechanically extracted from upstream sources.

One possible source of such assembly is the s2n-bignum project: https://github.com/awslabs/s2n-bignum

[spitters]

I wonder how it compares to Jasmin's https://github.com/haslab/libjbn

[tarcieri]

As a data point, the graviola crate has adopted assembly from s2n-bignum and seems to perform quite well: https://crates.io/crates/graviola