Learning-enhanced robust controller synthesis with rigorous statistical and control-theoretic guarantees