Revert ntree limit fix (#6616)

The old (before fix) best_ntree_limit ignores the num_class parameters, which is incorrect. In before we workarounded it in c++ layer to avoid possible breaking changes on other language bindings. But the Python interpretation stayed incorrect. The PR fixed that in Python to consider num_class, but didn't remove the old workaround, so tree calculation in predictor is incorrect, see PredictBatch in CPUPredictor.

tests/python/test_with_sklearn.py

@@ -106,7 +106,7 @@ def test_best_ntree_limit():
         )
 
         if forest:
-            assert cls.best_ntree_limit == rounds * forest * cls.n_classes_
+            assert cls.best_ntree_limit == rounds * forest
         else:
             assert cls.best_ntree_limit == 0