Towards a more robust treatment of discrete-time stochastic processes: developing a theoretical framework for working with imprecise probabilities in Markov chains