Rethinking Propagation for Unsupervised Graph Domain Adaptation