[crypto, otbn] Speeding up Montgomery constant calculations

Nobody has claimed this yet.

@JPulkus has proposed some ideas during our crypto meeting for speeding up the algorithm that we use to compute some constants for Montgomery arithmetic, and I think we can use this PR to have a more broader discussion on that.

I think it will take some time for me to get the full context here and also ramp up on OTBN, but here is my understanding so far:
* The so-called algorithm computes `R^2 mod P` and `P^{-1} mod R`, where `R = 2^256` and `P` is typically a large prime (group order of field prime).
* These two constants (as they are _constants_) do not need to be computed for every multiplication/addition/inversion. I would assume that we also compute these constants once at the beginning of a very long operation such as ECDSA signing/key generation/verification, which invokes many Montgomery multiplication+reduction operations. Therefore, this algorithm is not on the critical path in the sense that even significant speed-up improvements have small gain in the total speed-up.

I suppose if we decide to replace the existing algorithm, then we would probably want to have a written algorithm that is easier to understand and analyze (at least for the doc purposes). So, this issue can also be useful for that purpose.

I am self-assigning this, but I have some higher priority tasks in my task scheduler at the moment, so it might take me a while to actually start working on this. Also a related PR: #14944

Please feel free to correct me or add more context @jadephilipoom, @felixmiller, @johannheyszl, @JPulkus

Start with the Montgomery constant algorithm described in the issue and read related PR #14944 to recover the existing implementation and discussion context. The scope needs to be clarified before work begins: determine whether the goal is an algorithm replacement, a performance improvement, documentation, or some combination, then establish how completion will be assessed.

Written by the indexing model from the issue text.