# Description
While reviewing #10350 I noticed that a `HashSet<usize>` was used to
deduplicate the incoming rows, which are then sorted after cloning to a
separate `Vec`. This sounds like a candidate for a `BTreeSet` which
guarantees the ordering.
In the process I removed some dead code.
- Use `BTreeSet` instead of `HashSet`
- Remove dead `skip` logic
- Use `BTreeSet` directly in `NthIterator`
- Consume `BTreeSet` through `Peekable<IntoIter>`
1bb953877e