[gym] Ensure values passed to episode.hist_data are of native types

[mwulfman]

In python/gym_jiminy/rllib/gym_jiminy/rllib/callbacks.py, line 45:
episode.hist_data.setdefault(key, []).append(value)

if value is not of native type (i.e. for instance if it is numpy.float or numpy.int), it can create significant bottlenecks resulting in an important slow-down in training.

This is likely caused by Tensorboard.