I am of course aware of Austern's paper; it has seeded my mind with Ideas ever since you linked to it on these fora a decade or so ago. But it isn't my goal to faithfully port this work to Swift; my goal is to pervert it into something that aligns with the goals of Swift's roadmap for improving performance predictability -- i.e., the work that this effort is a part of.
I am taking the idea of a hierarchical iterator, and I'm deliberately misunderstanding it to mean exactly two levels, with simple contiguous buffers at the bottom. The problem at hand is a vexing performance issue; the solution I'm exploring is to significantly reduce abstraction, not to introduce more.
Sorry to disappoint!
(The pitch at hand is not about this though -- it is about the representation of those buffers.)