Check __cuda_array_interface__ instead of cupy class. (#9971)

* Now XGBoost can directly consume CUDA data from torch.

doc/python/python_intro.rst

@@ -162,6 +162,8 @@ Support Matrix
 +-------------------------+-----------+-------------------+-----------+-----------+--------------------+-------------+
 | cupy.ndarray            | T         | T                 | T         | T         | T                  | T           |
 +-------------------------+-----------+-------------------+-----------+-----------+--------------------+-------------+
+| torch.Tensor            | T         | T                 | T         | T         | T                  | T           |
++-------------------------+-----------+-------------------+-----------+-----------+--------------------+-------------+
 | dlpack                  | CPA       | CPA               |           | CPA       | FF                 | FF          |
 +-------------------------+-----------+-------------------+-----------+-----------+--------------------+-------------+
 | datatable.Frame         | T         | FF                |           | NPA       | FF                 |             |