Convergence of variance-reduced learning under random reshuffling