Reinforcement Learning in Presence of Discrete Markovian Context Evolution