Add reliable downstream dependents to nucleus

We added a basic nucleus config in #230, and chose which dependents we should test based on the results of a single run in #229. That's probably not an adequate sample size to identify which dependents can be reliably tested or which ones may be broken because of us. We should re-visit the choice of dependents, and ideally enable as many as possible.