Experiment with a native u128 type

[TomAFrench]

We currently simulate a u128 type using the U128 struct which causes a significant amount of overhead. We can try to avoid this by adding a new numeric type.

One reason why we don't have this currently is that we cannot safely multiply this type with itself as it may overflow the field modulus. We cannot then implement std::ops::arith::Mul for it.