Supervision versus Demonstration-Based In-Context Learning for Multiword Expression Classification