cudf.pandas: Series dt accessor is CombinedDatetimelikeProperties (#1…

…5523)

On the pandas Series type (not an instance) the dt attribute returns a CombinedDatetimelikeProperties object, which advertises the attributes of all possible datetime like dtypes. Previously we were proxying this with a DatatimeProperties object, which doesn't advertise as many properties. To allow wrapping libraries like dask that introspect the object to work correctly, advertise like pandas on the type. The instance still produces an object of the correct type due to dynamic lookup and/or metaclass magic in cudf.pandas and pandas respectively.

- Closes #15522

Authors:
  - Lawrence Mitchell (https://github.com/wence-)

Approvers:
  - Peter Andreas Entschev (https://github.com/pentschev)
  - GALI PREM SAGAR (https://github.com/galipremsagar)

URL: #15523

python/cudf/cudf/pandas/_wrappers/pandas.py

@@ -174,7 +174,7 @@ def _DataFrame__dir__(self):
         "__arrow_array__": arrow_array_method,
         "__cuda_array_interface__": cuda_array_interface,
         "__iter__": custom_iter,
-        "dt": _AccessorAttr(DatetimeProperties),
+        "dt": _AccessorAttr(CombinedDatetimelikeProperties),
         "str": _AccessorAttr(StringMethods),
         "cat": _AccessorAttr(_CategoricalAccessor),
         "_constructor": _FastSlowAttribute("_constructor"),
@@ -208,7 +208,7 @@ def Index__new__(cls, *args, **kwargs):
         "__array_function__": array_function_method,
         "__arrow_array__": arrow_array_method,
         "__cuda_array_interface__": cuda_array_interface,
-        "dt": _AccessorAttr(DatetimeProperties),
+        "dt": _AccessorAttr(CombinedDatetimelikeProperties),
         "str": _AccessorAttr(StringMethods),
         "cat": _AccessorAttr(_CategoricalAccessor),
         "__iter__": custom_iter,