The kleshni encode test is failing now because its implementation actually fails this same issue. They use a ~direct subtraction w/ a check for overflow as a tests of greater than. However this doesn't account for the structure of the field where an unreduced field element of value, say $2²⁵⁵ - 1$ (equivalent to 18) as greater than 30, for example.

So for test cases exist for kleshni where a "subtract and check for overflow" implementation of greater_than differs from a proper implementation.

This kind of shows that trying to accommodate a Legacy mode is not worth it. We don't even need an implementation of ct_gt() because it is actually just checking whether the FieldElement is negative or not, and we already have is_negative() for that. But is_negative() doesn't work for kleshni because they use the overflow of mpn_sub_n to check this.

Will decide how to proceed in the next commit.