Robust Covariate Control in Cluster-Randomized Trials