Investigate fearless_simd crate

[ounsworth]

Background

We will eventually have to look at SIMD to speed up bit of code that are performing the same computation many times in parallel.

The crate:
https://docs.rs/fearless_simd/latest/fearless_simd/
looks promising.

Task

• Perform a code review of the fearless_simd crate and form an opinion about whether this is a thing we would want as a core dependency. Some questions to think about:
  • Does it feel like it meets the same quality standards as bc-rust (code quality, readability, tests, etc).
  • Do we feel that if we absolutely had to, we have the skills and bandwidth to hard-fork and take over maintenance of that crate?
  • Is its license compatible with the above?
• Do a prototype. I would suggest that SIMD-vectorizing mldsa/polynomial.rs would give the most dramatic benchmark speedup of all the stuff that we've implemented so for.
• Think about build implications: would this need a new cargo feature, or could we safely leave it always-on?