secp256k1proto: split APrimeFE.from_bytes into checked and wrapping variants

[theStack]

Note that the issue title refers to a from_int method, but since the check-or-wraparound discussion mostly concerns the initialization of scalars/fes from byte-strings (e.g. pseudo-random data from tagged hash results), the split-up is applied on the .from_bytes method. As suggested in #77, the BIP-340 implementation is not adapted and still applies the wrapping manually (% GE.ORDER), in order to keep it close to the reference implementation.

Resolves #77.

[real-or-random]

Concept ACK

Given my comment in #80, perhaps we should do this:

• add from_int_checked and from_int_wrapping
• from_byte_checked and from_byte_wrapping can delegate to the from_int_* methods
• The constructor can be an alias for from_int_checked

[theStack]

The constructor can be an alias for from_int_checked

Hm, the APrimeFE constructor currently takes two parameters a,b for initializing numerator/denominator and we pass overflow values internally on a few places (see e.g. __add__, __mul__). Should we rename this constructor body to something like from_num_denom and have the actual constructor with one parameter instead? Or keep it as-is but throw if either of the values a,b are integers and overflow (with the drawback that the reduction has to be done on the call-sites internally)?

(Btw I plan to address this issue directly in secp256k1lab, but thought it makes sense to continue the discussion still here to not lose context).

[real-or-random]

Good point. So we shouldn't touch the constructor. And yes, we use it often within the class, so everything will break if it checks for overflow.

Perhaps we can document in the docstring of the class, or in the one of __init__, that it's is low-level and callers should usually prefer one of the factory methods.