[FEA] Add support for timezone-aware data to from_pandas

[charlesbluca]

Is your feature request related to a problem? Please describe.
Support for timezone-aware datetimes is ongoing in #12813, and currently timezone-aware operations like tz_convert and tz_localize are supported.

However, when trying to load in a pandas dataframe with timezone-aware data using from_pandas, we still error with a message that implies that timezone-aware datetimes aren't yet supported:

import cudf
import pandas as pd

df = pd.DataFrame(
    {
        "d": pd.date_range(
            start="2014-08-01 09:00", freq="8H", periods=6, tz="UTC"
        ),
    }
)

cudf.from_pandas(df)

NotImplementedError                       Traceback (most recent call last)
Cell In[1], line 12
      2 import pandas as pd
      4 df = pd.DataFrame(
      5     {
      6         "d": pd.date_range(
   (...)
      9     }
     10 )
---> 12 cudf.from_pandas(df)

File /datasets/charlesb/micromamba/envs/dask-sql-gpuci-py39/lib/python3.9/site-packages/nvtx/nvtx.py:101, in annotate.__call__.<locals>.inner(*args, **kwargs)
     98 @wraps(func)
     99 def inner(*args, **kwargs):
    100     libnvtx_push_range(self.attributes, self.domain.handle)
--> 101     result = func(*args, **kwargs)
    102     libnvtx_pop_range(self.domain.handle)
    103     return result

File /datasets/charlesb/micromamba/envs/dask-sql-gpuci-py39/lib/python3.9/site-packages/cudf/core/dataframe.py:7491, in from_pandas(obj, nan_as_null)
   7393 """
   7394 Convert certain Pandas objects into the cudf equivalent.
   7395 
   (...)
   7488 <class 'pandas.core.indexes.multi.MultiIndex'>
   7489 """
   7490 if isinstance(obj, pd.DataFrame):
-> 7491     return DataFrame.from_pandas(obj, nan_as_null=nan_as_null)
   7492 elif isinstance(obj, pd.Series):
   7493     return Series.from_pandas(obj, nan_as_null=nan_as_null)

File /datasets/charlesb/micromamba/envs/dask-sql-gpuci-py39/lib/python3.9/site-packages/nvtx/nvtx.py:101, in annotate.__call__.<locals>.inner(*args, **kwargs)
     98 @wraps(func)
     99 def inner(*args, **kwargs):
    100     libnvtx_push_range(self.attributes, self.domain.handle)
--> 101     result = func(*args, **kwargs)
    102     libnvtx_pop_range(self.domain.handle)
    103     return result

File /datasets/charlesb/micromamba/envs/dask-sql-gpuci-py39/lib/python3.9/site-packages/cudf/core/dataframe.py:5119, in DataFrame.from_pandas(cls, dataframe, nan_as_null)
   5115 for col_name, col_value in dataframe.items():
   5116     # necessary because multi-index can return multiple
   5117     # columns for a single key
   5118     if len(col_value.shape) == 1:
-> 5119         data[col_name] = column.as_column(
   5120             col_value.array, nan_as_null=nan_as_null
   5121         )
   5122     else:
   5123         vals = col_value.values.T

File /datasets/charlesb/micromamba/envs/dask-sql-gpuci-py39/lib/python3.9/site-packages/cudf/core/column/column.py:2327, in as_column(arbitrary, nan_as_null, dtype, length)
   2317         if cudf.get_option(
   2318             "default_float_bitwidth"
   2319         ) and infer_dtype(arbitrary) in (
   2320             "floating",
   2321             "mixed-integer-float",
   2322         ):
   2323             pa_type = np_to_pa_dtype(
   2324                 _maybe_convert_to_default_type("float")
   2325             )
-> 2327     data = as_column(
   2328         pa.array(
   2329             arbitrary,
   2330             type=pa_type,
   2331             from_pandas=True
   2332             if nan_as_null is None
   2333             else nan_as_null,
   2334         ),
   2335         dtype=dtype,
   2336         nan_as_null=nan_as_null,
   2337     )
   2338 except (pa.ArrowInvalid, pa.ArrowTypeError, TypeError):
   2339     if is_categorical_dtype(dtype):

File /datasets/charlesb/micromamba/envs/dask-sql-gpuci-py39/lib/python3.9/site-packages/cudf/core/column/column.py:1974, in as_column(arbitrary, nan_as_null, dtype, length)
   1968 if isinstance(arbitrary, pa.lib.HalfFloatArray):
   1969     raise NotImplementedError(
   1970         "Type casting from `float16` to `float32` is not "
   1971         "yet supported in pyarrow, see: "
   1972         "https://issues.apache.org/jira/browse/ARROW-3802"
   1973     )
-> 1974 col = ColumnBase.from_arrow(arbitrary)
   1976 if isinstance(arbitrary, pa.NullArray):
   1977     new_dtype = cudf.dtype(arbitrary.type.to_pandas_dtype())

File /datasets/charlesb/micromamba/envs/dask-sql-gpuci-py39/lib/python3.9/site-packages/cudf/core/column/column.py:340, in ColumnBase.from_arrow(cls, array)
    334 data = pa.table([array], [None])
    336 if (
    337     isinstance(array.type, pa.TimestampType)
    338     and array.type.tz is not None
    339 ):
--> 340     raise NotImplementedError(
    341         "cuDF does not yet support timezone-aware datetimes"
    342     )
    343 if isinstance(array.type, pa.DictionaryType):
    344     indices_table = pa.table(
    345         {
    346             "None": pa.chunked_array(
   (...)
    350         }
    351     )

NotImplementedError: cuDF does not yet support timezone-aware datetimes

Describe the solution you'd like
Either adding support for timezone-aware data to from_pandas, or updating the error message to indicate to users workarounds they can use while support is in progress.

Describe alternatives you've considered
It is generally possible to move timezone-aware data from pandas to cuDF by storing the timezone information somewhere, converting the data to timezone-naive (dt.tz_localize(None)) and then restoring the information once it's been read in with from_pandas.