Improve usage of polymorphism in columns (rapidsai#21030)

This PR reduces the amount of ColumnBase subclass-specific specialization present in method implementations in ColumnBase itself. It also completes some improvements around `__cuda_array_interface__` and mask setting. These improvements will as we work to establish a cleaner model for the ColumnBase<->pylibcudf.Column interop.

Authors:
  - Vyas Ramasubramani (https://github.com/vyasr)

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

URL: rapidsai#21030

Author: vyasr

python/cudf/cudf/core/column/categorical.py

@@ -38,6 +38,7 @@
         DtypeObj,
         ScalarLike,
     )
+    from cudf.core.buffer import Buffer
     from cudf.core.column import (
         ColumnBase,
         DatetimeColumn,
@@ -626,6 +627,23 @@ def copy(self, deep: bool = True) -> Self:
     def memory_usage(self) -> int:
         return self.categories.memory_usage + self.codes.memory_usage
 
+    @classmethod
+    def _deserialize_plc_column(
+        cls,
+        header: dict,
+        dtype: DtypeObj,
+        data: Buffer | None,
+        mask: Buffer | None,
+        children: list[ColumnBase],
+    ) -> plc.Column:
+        """Construct plc.Column from codes child for categorical columns.
+
+        Categorical columns store data as integer codes referencing categories.
+        The plc_column must be constructed from the codes child column, which
+        contains the actual integer data, rather than from the data/mask buffers.
+        """
+        return children.pop(0).plc_column
+
     @staticmethod
     def _concat(
         objs: MutableSequence[CategoricalColumn],