Variance-reduced stochastic learning by networked agents under random reshuffling