Improve secret independence guarantees in compiled code.

[xvzcf]

Following up on this: https://groups.google.com/a/list.nist.gov/g/pqc-forum/c/hqbtIGFKIpU, we should re-examine the way we write secret-independent code and we should also check the resulting binary code to see whether the compiler has optimized away from the constant time code or deviated in any other way from our intentions.

We could mark any function that is intended to produce a mask in constant time as "never inline".

We could also run checks on the resulting binary using https://github.com/rozbb/dudect-bencher. Perhaps we could mimic a TIMECOP like checking approach using Miri and https://doc.rust-lang.org/nomicon/leaking.html.