Add function to fix quasi-reduction

[yanesca]

Many optimised reduction routines don't give a canonical (least non-negative) representation, only reduce it to the appropriate limb size. We used to do this adjustment in the ecp module.

However according to our new semantic requirement the new Bignum routines need to return a canonical representative. The task is to add a function to ecp_curves that does this step and a unit test to test it:

• The function should take a raw pointer and a modulus as an input and reduce the value at the raw pointer input
• Valid inputs are the same bitlength as the modulus (Optimised reduction routines reduce the result - or will be reducing the result when the refactoring is complete - to the same bitlength as the modulus)
• Eventually we will want to make this constant time, the implementation should take this into account (eg _sub and then _add_if based on whether there was a carry).

Using this function from optimised reduction routines is out of scope for this task.

[gilles-peskine-arm]

Isn't this ecp_modp from ecp.c (transposed to the new types)? We always call grp->modp via this function.

[yanesca]

Yes, it serves the same purpose, but is simpler than ecp_modp(). With the new fixed width approach we know for sure that the result is <2p and that at most 1 subtraction is enough. ecp_modp() needs to loop, because there are more iterations needed.