`_is_pairwise` removed in scikit-learn 1.1 should be replaced by `_is_pairwise_metric`

[cmarmo]

What happened:

Checking scikit-learn#14687 the code throws the error

Traceback (most recent call last):
  File "/home/cmarmo/software/tests/14687-skl.py", line 2, in <module>
    import dask_ml.datasets
  File "/home/cmarmo/skldevenv/lib64/python3.10/site-packages/dask_ml/__init__.py", line 4, in <module>
    from dask_ml.model_selection import _normalize
  File "/home/cmarmo/skldevenv/lib64/python3.10/site-packages/dask_ml/model_selection/__init__.py", line 8, in <module>
    from ._search import GridSearchCV, RandomizedSearchCV, check_cv, compute_n_splits
  File "/home/cmarmo/skldevenv/lib64/python3.10/site-packages/dask_ml/model_selection/_search.py", line 18, in <module>
    from sklearn.base import (
ImportError: cannot import name '_is_pairwise' from 'sklearn.base' (/home/cmarmo/software/scikit-learn/sklearn/base.py)

This happens because _is_pairwise will be removed in scikit-learn 1.1 and should be replaced by _is_pairwise_metric : I have a pull request ready if welcomed.

What you expected to happen:
No import issues

Minimal Complete Verifiable Example:

# Put your MCVE code here
import dask.array as da
import dask_ml.datasets
import sklearn.linear_model

X, y = dask_ml.datasets.make_classification(chunks=50)

clf = sklearn.linear_model.LogisticRegression()
clf.fit(X, y)

Environment:

• Dask version: 1e811ce
• Python version: Python 3.10.2
• Operating System: Linux 5.16.13-200.fc35.x86_64
• Install method (conda, pip, source): source
• scikit-learn version: scikit-learn/scikit-learn@e5736af
• scikit-learn install method: source