[siddhantharsh]

Closes #425

Problem

MulticlassLabelDataset cannot be imported from aif360.datasets the way the other dataset classes can:

>>> from aif360.datasets import MulticlassLabelDataset
ImportError: cannot import name 'MulticlassLabelDataset' from 'aif360.datasets'

It is only reachable through aif360.datasets.multiclass_label_dataset, and it has no page in the API docs.

Fix

• Import the class in aif360/datasets/__init__.py, after BinaryLabelDataset. The order matters: multiclass_label_dataset.py does from aif360.datasets import StructuredDataset, so importing it before StructuredDataset is defined in the package raises a circular import error.
• Add it to "Base classes" in docs/source/modules/datasets.rst.
• Remove a stray zero-width space (U+200B) from the validate_dataset docstring in multiclass_label_dataset.py. It sat on what should be a blank line, and once the class was added to the docs it made Sphinx report "Unexpected indentation" on the class page. This is a one-line change and the file's CRLF line endings are unchanged.
• Add tests/test_multiclass_label_dataset.py.

No behaviour changes: this only makes an existing class importable from the package and documents it. No new dependencies or datasets.

Testing

Python 3.12, following CONTRIBUTING.

• The new tests fail on main with the ImportError above and pass with this change.
• pytest tests/test_multiclass_label_dataset.py tests/test_classification_metric.py tests/test_structured_dataset.py: 13 passed.
• Full pytest tests: no new failures compared with main. Every test that fails with this change also fails on main; in my environment those need R (MEPS, fairadapt) or hit dependency version drift (fmin_l_bfgs_b in the LFR tests, the bank loader).
• flake8 . --select=E9,F63,F7,F82 (the CI hard-fail rules): no errors.
• make html in docs/ builds, the MulticlassLabelDataset page is generated, and there are no warnings about it.

The commit is signed off (DCO).