Avoid cudf column APIs after cudf.Series disallows column inputs (#6019)

closes #6017

With rapidsai/cudf#16454, `cudf.Series` no longer accepts cudf `ColumnBase` objects. The usage of this pattern in cuml was mostly not needed as there existed public APIs to avoid this behavior

Authors:
  - Matthew Roeschke (https://github.com/mroeschke)

Approvers:
  - Bradley Dice (https://github.com/bdice)
  - Dante Gama Dessavre (https://github.com/dantegd)

URL: #6019

python/cuml/cuml/dask/preprocessing/LabelEncoder.py

@@ -114,8 +114,8 @@ class LabelEncoder(
         0    a
         1    a
         2    b
-        0    c
-        1    b
+        3    c
+        4    b
         dtype: object
         >>> client.close()
         >>> cluster.close()

python/cuml/cuml/preprocessing/LabelEncoder.py

@@ -217,10 +217,9 @@ def transform(self, y) -> cudf.Series:
 
         y = cudf.Series(y, dtype="category")
 
-        encoded = y.cat.set_categories(self.classes_)._column.codes
-        encoded = cudf.Series(encoded, index=y.index)
+        encoded = y.cat.set_categories(self.classes_).cat.codes
 
-        if encoded.has_nulls and self.handle_unknown == "error":
+        if encoded.hasnans and self.handle_unknown == "error":
             raise KeyError("Attempted to encode unseen key")
 
         return encoded
@@ -237,9 +236,9 @@ def fit_transform(self, y, z=None) -> cudf.Series:
         self.dtype = y.dtype if y.dtype != cp.dtype("O") else str
 
         y = y.astype("category")
-        self.classes_ = y._column.categories
+        self.classes_ = y.cat.categories
 
-        return cudf.Series(y._column.codes, index=y.index)
+        return y.cat.codes
 
     def inverse_transform(self, y: cudf.Series) -> cudf.Series:
         """
@@ -275,11 +274,14 @@ def inverse_transform(self, y: cudf.Series) -> cudf.Series:
 
         y = y.astype(self.dtype)
 
-        ran_idx = cudf.Series(cp.arange(len(self.classes_))).astype(self.dtype)
-
-        reverted = y._column.find_and_replace(ran_idx, self.classes_, False)
+        # TODO: Remove ._column once .replace correctly accepts cudf.Index
+        ran_idx = (
+            cudf.Index(cp.arange(len(self.classes_)))
+            .astype(self.dtype)
+            ._column
+        )
+        res = y.replace(ran_idx, self.classes_)
 
-        res = cudf.Series(reverted)
         return res
 
     def get_param_names(self):