provide option for automatic dump of (state or extended state, action, reward) trajectories

[jerabaul29]

To enable a posteriori analysis / ROM based on the trajectories / model based RL, it could be useful to dump the (state or extended state, action, reward) trajectories. What I mean by "extended state" is that the state dumped could even be a state larger than what the ANN actually sees, to enable extra analysis.

Having a config file to set this up / allow to configure dumping and paths and quantity of data dumped etc would be great :) .