`xp.isdtype` checks raise on NumPy `StringDType`

[!WARNING] This issue is not yet ready for a PR. If you are interested in contributing to scikit-learn, please have a look at our contributing guidelines, and in particular the sections for new contributors and the "Needs triage" label.

Describe the bug and give evidence about its user-facing impact

numpy.dtypes.StringDType (and other new-style DType classes) crash xp.isdtype() across many sklearn utilities.

numpy's (and array-api-compat's) isdtype() raises TypeError instead of returning False for numpy's new-style DType classes. sklearn calls xp.isdtype(...) in many places expecting False for unrecognized dtypes, not a crash.

Steps/Code to Reproduce

import numpy as np
np.isdtype(np.dtypes.StringDType(), "numeric")
# TypeError: dtype argument must be a NumPy dtype, but it is a <class 'numpy.dtypes.StringDType'>.

from sklearn.utils.multiclass import type_of_target
type_of_target(np.array(["a", "b", "a"], dtype=np.dtypes.StringDType()))
# TypeError: dtype argument must be a NumPy dtype, but it is a <class 'numpy.dtypes.StringDType'>.

Confirmed crash sites (14), all via xp.isdtype() except unique_labels:

Function	File
check_array (any dtype=, incl. None)	utils/validation.py
type_of_target	utils/multiclass.py
_is_integral_float	utils/multiclass.py
_assert_all_finite	utils/validation.py
_randomized_range_finder	utils/extmath.py
_safe_accumulator_op	utils/extmath.py
_find_matching_floating_dtype	utils/_array_api.py
_average	utils/_array_api.py
label_binarize	preprocessing/_label.py
LabelBinarizer.inverse_transform	preprocessing/_label.py
RidgeClassifier.fit	linear_model/_ridge.py
_encode	utils/_encode.py
_check_unknown (→ OrdinalEncoder.fit_transform)	utils/_encode.py
unique_labels (np.dtype(y.dtype, metadata=...))	utils/_unique.py

Not affected: is_multilabel, LabelEncoder.fit, _array_indexing, _determine_key_type.

Related: #34383 (motivation — proposes StringDType internally in encoders), #9777 (same failure pattern, closed).

Expected Results

Unrecognized dtypes should be treated like any other unsupported dtype (e.g. return False/"unknown"), not raise.

Actual Results

TypeError at every site listed above (unique_labels raises via a different call, same root cause).

Versions

Python 3.11.15
scikit-learn 1.8.0 and 1.9.1 (both confirmed broken)
numpy 2.4.4
pandas 3.0.2
polars 1.44.0

Interest in fixing the bug

One internal helper, e.g. _safe_isdtype(), wrapping xp.isdtype() to catch TypeError and return False, used at all sites above instead of patching each individually.