Restore hashability of Extension objects (broken since 84.0.0 dataclass conversion)

[aryansk]

Fixes #5303

Problem

Since v84.0.0, Extension instances are unhashable:

>>> from setuptools.extension import Extension
>>> hash(Extension('x', ['y.c']))
TypeError: unhashable type: 'Extension'

The dataclass conversion of the vendored distutils.extension.Extension (pypa/distutils#373) used the default eq=True, which generates __eq__ and therefore sets __hash__ = None. Beyond the TypeError, this also silently changed equality semantics: two distinct Extension instances with equal field values now compare equal, so ext in some_list can be true for an object that was never appended.

Change

Pass eq=False to lenient_dataclass in setuptools/_distutils/extension.py, restoring the identity-based equality and hashability Extension always had before the conversion. The lenient unknown-keyword behaviour is unchanged.

Note: the same conversion exists upstream in pypa/distutils, so the vendored copy will need this fix re-applied at the next sync unless it lands there first — happy to send that PR if preferred.

Validation

• Reproduced on clean main (hash() → TypeError, Extension.__hash__ is None) with CPython 3.13; after the change: hashable, distinct instances unequal, usable as set members/dict keys.
• Lenient behaviour preserved: unknown keyword arguments still emit the deprecation warning.
• Added a regression test (test_extension_is_hashable); setuptools/_distutils/tests/test_extension.py: 4 passed, 2 skipped (optional mypy inference checks).
• News fragment added (newsfragments/5303.bugfix.rst).

[abravalheri]

Thank you for the contribution @aryansk .

Please have a look on https://setuptools.pypa.io/en/latest/development/developer-guide.html#code-conventions-and-other-practices regarding the _distutils folder.

There is already pypa/distutils#427 pending of review in the pypa/distutils repo.

[aryansk]

Thank you for the pointer @abravalheri — you're right that _distutils changes belong upstream. I see pypa/distutils#427 already covers this exact regression (#426), so closing this draft as superseded. Thanks for the quick review.