Add support for lightning.pytorch and extend optional dependencies for GraphGym

[jherasdo]

Dear PyG Team,

This PR introduces compatibility with the new lightning package (formerly pytorch_lightning), which is now the actively maintained version of the PyTorch Lightning framework.

Changes Included:

• Adds fallback import logic to support both lightning.pytorch and the legacy pytorch_lightning, preserving backward compatibility.
• Updates the graph_gym as-is with pytorch-lightning dependency.
• Introducing a new optional group graphgym-lightning for users adopting the newer lightning package.

Motivation & Benefits:

• Ensures GraphGym modules remain functional with current and future Lightning versions.
• Maintains support for users with legacy setups.
• Aligns with Lightning AI's latest rebranding and package structure.

Validation:
This compatibility approach has been tested on my own research projects, including multi-modal deep learning models built with PyTorch Geometric and Lightning. The fallback logic and both dependency setups work reliably in real-world scientific workflows.

Let me know if you'd like me to add tests or docs for this update - Happy to help!