On the performance of random reshuffling in stochastic learning