Migrate from OpenAI Gym to Farama Foundation Gymnasium

[StephenOman]

Breaking Changes

Migration from OpenAI Gym to the Farama Foundation Gymnasium introduces breaking modifications to some API calls.

For details, see v21 to v26 Migration Guide on the Farama Foundation website.

Main changes:

• Seeding can only take place on reset()
• reset() also allows options to be passed into the environment
• reset() returns additional information
• step() returns an additional value to indicate if the episode was truncated as the fourth item in the tuple
• render() no longer accepts the rendering mode as a parameter. Instead, rendering mode should be initialised when the environment is created in gym.make()

This PR also includes documentation changes and some minor bug fixes.

[StephenOman]

When this is released, a small change is needed to test_package.yml:31:

python -c 'import nle; import gymnasium as gym; e = gym.make("NetHack-v0"); e.reset(); e.step(0)'

[heiner]

Thanks! I have not tested this but the changes LGTM.

Once this landed and we updated the other test, we should also bump the version (to 0.9.3, or if you're daring to 1.0?) and then do a release to PyPI.

[StephenOman]

Once this landed and we updated the other test, we should also bump the version (to 0.9.3, or if you're daring to 1.0?) and then do a release to PyPI.

I definitely vote for a major version change since there are breaking changes in the API. Given the change to new repo etc., 1.0.0 is perfect.

[heiner]

1.0.0 sgtm.