Fix add_column parameter order so feature precedes new_fingerprint

[gaurav0107]

What

Dataset.add_column's signature is (name, column, new_fingerprint=None, feature=None), but:

• its own docstring documents the third argument as feature (not new_fingerprint), and
• every sibling @fingerprint_transform method keeps new_fingerprint as the last parameter (e.g. cast_column(column, feature, new_fingerprint=None), rename_column, remove_columns, select_columns).

Because new_fingerprint sits before feature, a docstring-conformant positional call raises:

>>> from datasets import Dataset, Value
>>> ds = Dataset.from_dict({"a": [1, 2]})
>>> ds.add_column("b", [3, 4], Value("int32"))
TypeError: Dataset.add_column() got multiple values for argument 'new_fingerprint'

The fingerprint_transform decorator injects new_fingerprint as a keyword argument, so the positional value collides with it.

Fix

Reorder the signature to (name, column, feature=None, new_fingerprint=None), matching the docstring and the cast_column convention. The decorator references new_fingerprint by keyword, so this is safe for existing keyword callers, and no Dataset.add_column caller in the repo passes feature/new_fingerprint positionally.

Test

Added test_add_column_feature_argument in tests/test_arrow_dataset.py, asserting that a positional feature is honored, the keyword form still works, and the new_fingerprint keyword is still respected. The test fails on main (TypeError) and passes with this change.

Fixes #7864

---

🤖 Opened with Superhuman, an open-source contribution agent.