Approximate Probabilistic Constraints and Risk-Sensitive Optimization Criteria in Markov Decision Processes