gym.Env.reset() no longer returns observation of type np.array but a tuple of (observation, info)

[Aldair Cuarez]

A recent update to OpenAI Gym changed the way envs behave.
gym.Env.reset() now returns a tuple of (observation, info) so I think that every instance of observation = deepcopy(env.reset()) in keras-rl.rl.core.Agent.fit() should be change to observation, info = deepcopy(env.reset()).