Uniform Positive Recurrence And Long Term Behavior Of Markov Decision Processes, With Applications In Sensor Scheduling