Associated types need to do transitive bound expansion

[nikomatsakis]

Today when you create a function that has types bounded by a trait with associated types, you get implicit parameters corresponding to those associated types. But those type parameters themselves may have bounds (#18178) which may in turn have associated types, so we really ought to expand to the transitive closure. (Or, better yet, find another impl technique, since the transitive closure does not necessarily have finite size if you have a trait that uses itself as a bound, but we can accept this limitation for now).

[nikomatsakis]

No longer relevant in the new associated types implementation.