Tests fail with pandas 3

When preparing Fedora for upgrading to pandas3 the following tests fail when building against the new version. Would it be possible to port to support the new version.

```
=================================== FAILURES ===================================
______________________________ test_PandasColumns ______________________________

def test_PandasColumns():
with PandasColumns() as p:
assert os.path.exists(p.partd.partd.path)

> p.append({'x': df1, 'y': df2})

partd/tests/test_pandas.py:32:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
../BUILDROOT/usr/lib/python3.14/site-packages/partd/pandas.py:56: in append
self.partd.append(arrays, **kwargs)
../BUILDROOT/usr/lib/python3.14/site-packages/partd/numpy.py:55: in append
self.partd.iset(suffix(k, '.dtype'), serialize_dtype(v.dtype))
^^^^^^^^^^^^^^^^^^^^^^^^
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

dt =

def serialize_dtype(dt):
""" Serialize dtype to bytes

>>> serialize_dtype(np.dtype('i4'))
b'>> serialize_dtype(np.dtype('M8[us]'))
b' return dt.str.encode()
^^^^^^
E AttributeError: 'StringDtype' object has no attribute 'str'

../BUILDROOT/usr/lib/python3.14/site-packages/partd/numpy.py:25: AttributeError
____________________________ test_column_selection _____________________________

def test_column_selection():
with PandasColumns('foo') as p:
> p.append({'x': df1, 'y': df2})

partd/tests/test_pandas.py:51:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
../BUILDROOT/usr/lib/python3.14/site-packages/partd/pandas.py:56: in append
self.partd.append(arrays, **kwargs)
../BUILDROOT/usr/lib/python3.14/site-packages/partd/numpy.py:55: in append
self.partd.iset(suffix(k, '.dtype'), serialize_dtype(v.dtype))
^^^^^^^^^^^^^^^^^^^^^^^^
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

dt =

def serialize_dtype(dt):
""" Serialize dtype to bytes

>>> serialize_dtype(np.dtype('i4'))
b'>> serialize_dtype(np.dtype('M8[us]'))
b' return dt.str.encode()
^^^^^^
E AttributeError: 'StringDtype' object has no attribute 'str'

../BUILDROOT/usr/lib/python3.14/site-packages/partd/numpy.py:25: AttributeError
=========================== short test summary info ============================
FAILED partd/tests/test_pandas.py::test_PandasColumns - AttributeError: 'Stri...
FAILED partd/tests/test_pandas.py::test_column_selection - AttributeError: 'S...
=================== 2 failed, 53 passed, 1 skipped in 1.16s ====================
```

Start with partd/numpy.py, where serialize_dtype raises the StringDtype error, and partd/tests/test_pandas.py, especially test_PandasColumns and test_column_selection. Reproduce the failures with pandas 3, then verify those tests and the existing suite pass without the reported AttributeError.

Written by the indexing model from the issue text.