Fix multiplication in Uint64::multiply_ratio and Uint128::multiply_ratio

[webmaster128]

Right now we use an implementation like this:

        // TODO: minimize rounding that takes place (using gcd algorithm)
        let val = self.u64() * numerator / denominator;
        Uint64::from(val)

which is executed as (self.u64() * numerator) / denominator. This means that as soon as the multiplication overflows, the execution panics. This is an undesired behaviour because all of those:

base.multiply_ratio(1u64, 1u64);
base.multiply_ratio(3u64, 3u64);
base.multiply_ratio(u64::MAX, u64::MAX)

should return the same value.

In order to make this happen, we need to support a full multiplication that cannot overflow. For the Uint64 case this means multiplying two u64s into a u128 and for the Uint128 case this means multiplying two u128s into a 256 bit integer. The former can be done by casting to u128. The later is implemented e.g. in paritytech/parity-common#546. Maybe we can implement it manually though.