`Ord:cmp` does not properly handle `Ratio<T>` with negative denominators

[mikem8891]

I am not sure if Ratio is supposed to handle negative denominators properly, but the Ord::cmp function does not, see the example below.

use num_rational::Ratio; // 0.4.2

fn main() {
    let neg_half = Ratio::<i8>::new_raw(1, -2);
    let one_third = Ratio::<i8>::new_raw(1, 3);
    println!("{neg_half} is {:?} than {one_third}", neg_half.cmp(&one_third));
    println!("{one_third} is {:?} than {neg_half}", one_third.cmp(&neg_half));
}

results in

1/-2 is Greater than 1/3
1/3 is Less than 1/-2

[cuviper]

I'm getting ready to bump semver, if only to raise the num-bigint dependency, but I'm thinking this may be a good time to disavow negative denominators. Then we could add debug asserts in strategic places and outright remove any code that's currently trying to cope with negative denominators. What do you think?

(Non-reduced ratios are also annoying, but that's more expensive to verify and/or fix, so I'm more willing to accommodate that.)

[mikem8891]

I'm thinking this may be a good time to disavow negative denominators.

Yes, please! It will need to be made clear that positive denominators are an invariant and that the user is expected to enforce it when using Ratio::new_raw.

Do you want me to change #147 to eliminate the denominator check and add a debug assert?