refactor: improve consistency of extras

[clement-bonnet]

Is your feature request related to a problem? Please describe

The extras field of TimeStep can contain environment information useful for decision-making (e.g. Connect4's current player ID) or environment metrics (e.g. BinPack's volume utilisation). There is an inconsistency in what the extras field is used for as it is sometimes meant to be used by the algorithm and sometimes just logged as a metric.

Describe the solution you'd like

We should move any algorithm-related information from extras to the environment observation (e.g. Connect4's observation could have another field called current_player or something). We should update the documentation/docstrings accordingly to explicitly mention that TimeStep.extras does not contain stuff that is meant to be observed by the agent as those should be in the observation.

---

TODOs

• adapt docstrings, doc, codes, etc to make explicit the fact that TimeStep.extras does not contain any info meant to be observed
• move agent-specific extras (e.g. Connect4's current player ID) to environment observations

[clement-bonnet]

It has been discussed that keeping extras as naming is better than forcing metrics. I am changing the issue description accordingly.