If multiple indices contain the same version, use the first index

[yorickvP]

This fixes resolving packages that publish an invalid stub to pypi, such as tensorrt-llm.

Summary

In #3138 , we implemented unsafe-best-match. However, it seems to not quite work as expected. When multiple indices contain the same version, it's not clear which index the current code uses. This PR fixes that to use the first index the package is in.

Test Plan

$ echo 'tensorrt-llm==0.11.0' | ./target/debug/uv pip compile - --extra-index-url https://pypi.nvidia.com --python-version=3.10 --index-strategy=unsafe-best-match --annotation-style=line

cc: @charliermarsh

[charliermarsh]

Yeah interesting, contrary to my expectation kmerge isn't "stable" by iterator:

#[cfg(test)]
mod tests {
    use itertools::Itertools;

    #[test]
    fn kmerge_reverse() {
        let a = vec![('a', 1), ('b', 2), ('c', 3)];
        let b = vec![('d', 1), ('e', 2), ('f', 3)];

        let its = [a, b];
        let its = its.iter().kmerge_by(|a, b| a.1 < b.1);

        let actual = its.collect::<Vec<_>>();
        let expected = vec![&('a', 1), &('d', 1), &('b', 2), &('e', 2), &('c', 3), &('f', 3)];

        // This fails; left is: `[('a', 1), ('d', 1), ('e', 2), ('b', 2), ('c', 3), ('f', 3)]`
        assert_eq!(actual, expected);
    }
}

I didn't have any basis for expecting that so I don't think it's a bug, I just assumed it was true.

[charliermarsh]

@yorickvP - I tweaked to avoid re-comparing. Can you test that it's still passing on your test case?

[yorickvP]

Still works!

[charliermarsh]

Thanks!