Randomization-based causal inference from split-plot designs