Skops needs to set trusted types to deserialize sklearn KNN model

[WeichenXu123]

Skops needs to set trusted types to deserialize sklearn KNN model:

Repro code:

import skops.io
import sklearn.neighbors as knn
from sklearn import datasets
import pandas as pd

iris = datasets.load_iris()
X = iris.data
y = iris.target
X_df = pd.DataFrame(X, columns=iris.feature_names)
X_df = X_df.iloc[:, :2]  # we only take the first two features.
y_series = pd.Series(y)

knn_model = knn.KNeighborsClassifier()
knn_model.fit(X_df, y_series)

save_path = "/tmp/knn_model1.sko"
skops.io.dump(knn_model, save_path)

# If removing `trusted` param, loading fails.
# but the 2 types are sklearn builtin types, so that it should not ask user to set `trusted` param
skops.io.load(save_path, trusted=["sklearn.metrics._dist_metrics.EuclideanDistance64","sklearn.neighbors._kd_tree.KDTree"])

The 2 types ["sklearn.metrics._dist_metrics.EuclideanDistance64","sklearn.neighbors._kd_tree.KDTree"] are sklearn builtin types,
so that it should not ask user to set trusted param.

[adrinjalali]

I wouldn't say this is a bug. We have a limited set of trusted types by default, and we'd only add them after checking the code of the corresponding classes.

In general, the idea is not that users can load models w/o checking trusted types, and in mlflow we should also make sure they have a way to do that, also ref mlflow/mlflow#18889