Simple and Effective Graph Autoencoders with One-Hop Linear Models