Continuous Action Support and Environment/CI Improvement

[misaghsoltani]

This pull request adds support for continuous and large action spaces in DeepXube by adding a new domain (cont_opt), a lightweight MLP heuristic (cont_mlp), and support for Q* search (bwqs_optim). It also adds Pixi environment management to the project and updates the CI workflow for better cross-platform support. Below are the most important changes:

Continuous Action Space Support:

• Added ActsOptim mixin to deepxube/base/domain.py for domains that select actions via optimization, enabling action selection in continuous or large action spaces.
• Added deepxube/domains/cont_opt.py with the ContOptDomain class, supporting continuous state, goal, and action representations, and providing action proposals.

Heuristic and Pathfinding Enhancements:

• Added deepxube/heuristics/cont_mlp.py, implementing a lightweight MLP heuristic for continuous domains with flattened numeric inputs.
• Added deepxube/pathfinding/bwqs_optim.py, a batch weighted Q* search variant that leverages domain-dependant action proposals for search in large/continuous action spaces.

Environment and CI Improvements:

• Added pixi.toml and .gitattributes to support Pixi-based environment management, multi-platform support, and proper file handling. [1] [2]
• Updated .github/workflows/test.yml to use Pixi, add Windows testing, and some minor updates.
• Updated Python optional dependencies in pyproject.toml to include tox-uv for faster test environments.
• Deprecated the numpy.typing.mypy_plugin in mypy.ini per NumPy 2.3 recommendations.