Add JaxARC environment setup and documentation

[aadimator]

This pull request adds support for the JaxARC environment, enabling massively parallel RL training on ARC (Abstraction and Reasoning Corpus) puzzle tasks using JAX. It introduces new environment configuration files, updates dependencies, provides comprehensive documentation, and integrates JaxARC into the environment creation pipeline. These changes make it easy to run and customize ARC RL experiments in Stoix using JaxARC.

JaxARC integration and environment support:

• Added a new function make_jaxarc_env to stoix/utils/make_env.py for initializing and wrapping JaxARC environments, including support for domain-specific metrics via the ExtendedMetrics wrapper.
• Registered jaxarc in the environment factory mapping, enabling selection of JaxARC environments via config.

Configuration and customization:

• Added four new environment config files for JaxARC: default.yaml, agi1_train.yaml, agi1_bbox.yaml, and concept.yaml, covering MiniARC, ARC-AGI-1, bounding-box actions, and concept-based tasks. [1] [2] [3] [4]

Documentation:

• Created docs/envs/jaxarc.md, a detailed guide covering JaxARC usage, available datasets, action modes, observation wrappers, and key metrics.

Dependencies:

• Added jaxarc>=1.0.3 to the project dependencies in pyproject.toml.