Description bounds could have type `Antichain`.

In several places we use Vec<T> and &[T] to indicate frontiers, with the implied contract that they should be antichains. We could actually use antichains through timely dataflow's Antichain<T> type.

Perhaps with some additional wrapping (e.g. Frontier<T>) we (or timely) could partially order these antichains too, using the rule that one comes before another if each element of the other comes after some element of the former.

fn le(a1: &Frontier<T>, a2: &Frontier<T>) -> bool {
    a2.elements().iter().all(|t2| a1.elements().iter().any(|t1| t1.le(t2)))
}

This would clean up a bit of code (not lots) and make parts less error-prone / higher confidence.

The issue names no files or tests. Start by locating uses of Vec and &[T] that represent frontiers, then compare those contracts with timely dataflow's Antichain. Determine whether a Frontier wrapper and the proposed partial ordering are appropriate, and verify that the affected code becomes less error-prone.