How should domain models be designed — rich domain models with encapsulated logic vs. anemic models with separate service/util layers?

From a software architecture and clean code perspective

If by "clean code" you mean something closely aligned with the writings of Robert Martin... Eric Evans (Domain Driven Design) comes from that same tradition (OO, pattern languages, etc); if you compare the models that Evans uses for describing container shipping and the models that Martin uses for describing coffee makers, you'll doubtless notice the similarities.

On the other hand, if by "clean code" you mean easy to maintain / easy to modify / easy to delete... that can be achieved in any of the styles you list in the hands of a programmer experienced enough to understand the trade offs.

(Likewise, all three styles can turn into a mess in the hands of someone who doesn't understand the trade offs).