Support _estimator_type. (#6582)

* Use `_estimator_type`.

For more info, see: https://scikit-learn.org/stable/developers/develop.html#estimator-types

* Model trained from dask can be loaded by single node skl interface.

tests/python/test_with_sklearn.py

@@ -1099,3 +1099,26 @@ def test_boost_from_prediction_approx():
 @pytest.mark.skipif(**tm.no_sklearn())
 def test_boost_from_prediction_exact():
     run_boost_from_prediction('exact')
+
+
+def test_estimator_type():
+    assert xgb.XGBClassifier._estimator_type == "classifier"
+    assert xgb.XGBRFClassifier._estimator_type == "classifier"
+    assert xgb.XGBRegressor._estimator_type == "regressor"
+    assert xgb.XGBRFRegressor._estimator_type == "regressor"
+    assert xgb.XGBRanker._estimator_type == "ranker"
+
+    from sklearn.datasets import load_digits
+
+    X, y = load_digits(n_class=2, return_X_y=True)
+    cls = xgb.XGBClassifier(n_estimators=2).fit(X, y)
+    with tempfile.TemporaryDirectory() as tmpdir:
+        path = os.path.join(tmpdir, "cls.json")
+        cls.save_model(path)
+
+        reg = xgb.XGBRegressor()
+        with pytest.raises(TypeError):
+            reg.load_model(path)
+
+        cls = xgb.XGBClassifier()
+        cls.load_model(path)  # no error