Refactor: Generalize dataset base classes & consolidate dynamic splitting logic

[aditya0by0]

Description:
Currently, the code related to dynamic splitting in chebi.py and the proteins repo’s data class is duplicated. Both implementations are effectively the same, which leads to unnecessary code redundancy.

Proposed changes:

1. Move common code to base class — e.g., DynamicDataset — to encapsulate shared dynamic splitting logic.

   • Both ChEBI and protein dataset classes should inherit from this base class.
   • This will centralize changes and make maintenance easier.

2. Refactor dataset hierarchy to be more generic:

   • Certain hyperparameters that are specific to ChEBI, such as

     chebi_version: int = 200

     in XYBaseDataModule, should be pushed down into a ChEBI-specific base class rather than existing in a generic base.

3. Outcome:

   • Eliminate duplicate code between chebi.py and the proteins repo.
   • Improve maintainability by isolating dataset-specific configurations.
   • Make it easier to introduce new datasets without rewriting the splitting logic.