Recurrent, nonequilibrium systems and the Markov blanket assumption