Chaos from nonlinear Markov processes: Why the whole is different from the sum of its parts