Add RNN policy to `sb3-contrib`

[gcroci2]

sb3-contrib currently lacks support for simple RNN policies. To address this, we can adapt the existing LSTM-based policy implementation.

Although this is outside the scope of the NeuroGym library itself, we need this functionality for another project that builds on NeuroGym's features. I propose forking the original repository under the neurogym organization and creating a pull request to the upstream repo. This would be a valuable addition to the library and could benefit other users as well.

[gcroci2]

I’ve opened a pull request to the original stable-baselines3-contrib repository, which is currently pending review by the maintainers.

In the meantime, the feature is already available via our fork under the NeuroGym organization: https://github.com/neurogym/stable-baselines3-contrib (branch: rnn_policy_addition), and can be installed directly for immediate use.