optim: avoid div in millerLoopAndFinalExpResult

[shramee]

Description

Super minor optimisation to replace a div with a mul, for exponentiation by λ' = q^3 - q^2 + q. We already have the inv so we can just exp the inverse by q^2 and mul (instead of exp and div).

How has this been tested?

• Ran all the tests and benches

[ivokub]

cc @yelhousni , see also #1347 (reply in thread)

[yelhousni]

Thanks for spotting it 👍
Indeed it saves 1585 scs corresponding to the equality check in the hinted DivUnchecked.