[Feature Request] Add IMED-RL agent

[riiswa]

Hey,

I will add the experimental agent IMED-RL from the paper
IMED-RL: Regret optimal learning of ergodic Markov decision processes, Fabien Pesquerel, Odalric-Ambrym Maillard

This is a learning policy that is asymptotically optimal with respect to regret minimization problem under the average-reward criterion in ergodic MDPs with unknown reward and transition.