Randomization inference for balanced cluster-randomized trials