Support __len__ in ensemble estimators (#6468)

`sklearn` ensemble estimators are valid sequences of estimators. Supporting `__getitem__` and `__iter__` is _hard_ with our current implementation, but `__len__` is easy and lets more of the sklearn compatiblity tests pass.

Fixes #6465.

Authors:
  - Jim Crist-Harif (https://github.com/jcrist)

Approvers:
  - Simon Adorf (https://github.com/csadorf)

URL: #6468

Author: jcrist

python/cuml/cuml/ensemble/randomforest_common.pyx

@@ -192,6 +192,10 @@ class BaseRandomForestModel(UniversalBase):
         self.treelite_serialized_model = None
         self._cpu_model_class_lock = threading.RLock()
 
+    def __len__(self):
+        """Return the number of estimators in the ensemble."""
+        return self.n_estimators
+
     def _get_max_feat_val(self) -> float:
         if isinstance(self.max_features, int):
             return self.max_features/self.n_cols

python/cuml/cuml/tests/test_random_forest.py

@@ -1447,3 +1447,10 @@ def test_rf_predict_returns_int():
     clf = cuml.ensemble.RandomForestClassifier().fit(X, y)
     pred = clf.predict(X)
     assert pred.dtype == np.int64
+
+
+def test_ensemble_estimator_length():
+    X, y = make_classification()
+    clf = cuml.ensemble.RandomForestClassifier(n_estimators=3)
+    clf.fit(X, y)
+    assert len(clf) == 3