DQN rlax + bsuite vs rlax + gymnax

[Robert Tjarko Lange]

I would like to have a benchmark figure comparing the DQN example in rlax with a gymnax sped up version. Ideally, I want to compare the runtime for step transitions on different devices.

At the moment there is something wrong with the optimisation and/or evaluation. Figure out the bug 🐛.

The agents should all be in an experimental directory.

[Robert Tjarko Lange]

• Alternatively/additionally we can simply drop in for the Anakin Catch Example.
• Also add the CMA-ES example for Pendulum-v0 as a notebook!

[Robert Tjarko Lange]

Addressed in d7e262b and 9b92dbe.