Add a few simplifications around Mont inverse

[tcoratger]

Description

• When using div2 method of BigInt, we traverse the members in the opposite direction (big endian order) with a loop which until now was carried out in the little endian order. Therefore, in each iteration of the loop, the calculation of N - i - 1 was necessary. If we go through the limbs in the opposite direction with a while as I propose here, the calculation is no longer carried out on each pass, because we take the member i directly.
• In the inverse method of the Montgomery back end, when we exit the following piece of code:

if a.is_zero() {
  None
}

there is no need for the else, because the rest will necessarily take place if a is not zero. Therefore, I removed the else here to avoid unnecessary indentations.