BUG: CUDF deserialize error when data contains NA type like np.nan or pd.NA

[ChengjieLi28]

Note that the issue tracker is NOT the place for general support. For discussions about development, questions about usage, or any general questions, contact us on https://discuss.xorbits.io/.

Minimal Reproduction：

df = md.DataFrame({'a': ["foo", "baz", np.nan]}, chunk_size=1).to_gpu()
res = df.execute().fetch(to_cpu=False)
print(res)

Error:

_mars/dataframe/core.py:845: in fetch
    batches = list(self._iter(batch_size=batch_size, session=session, **kw))
_mars/dataframe/core.py:829: in _iter
    yield self._fetch(session=session, **kw)
_mars/core/entity/executable.py:169: in _fetch
    return fetch(self, session=session, **kw)
_mars/deploy/oscar/session.py:1760: in fetch
    return session.fetch(tileable, *tileables, **kwargs)
_mars/deploy/oscar/session.py:1562: in fetch
    return asyncio.run_coroutine_threadsafe(coro, self._loop).result()
/home/lichengjie/miniconda3/lib/python3.9/concurrent/futures/_base.py:446: in result
    return self.__get_result()
/home/lichengjie/miniconda3/lib/python3.9/concurrent/futures/_base.py:391: in __get_result
    raise self._exception
_mars/deploy/oscar/session.py:1728: in _fetch
    data = await session.fetch(tileable, *tileables, **kwargs)
_mars/deploy/oscar/tests/session.py:64: in fetch
    results = await super().fetch(*tileables, **kwargs)
_mars/deploy/oscar/session.py:1019: in fetch
    fetched_data = await fetcher.get()
_mars/services/task/execution/mars/fetcher.py:60: in get
    fetched_data = await storage_api.get.batch(
/home/lichengjie/miniconda3/lib/python3.9/site-packages/xoscar/batch.py:149: in _async_batch
    return await self.batch_func(args_list, kwargs_list)
_mars/services/storage/api/oscar.py:113: in batch_get
    return await self._storage_handler_ref.get.batch(*gets)
/home/lichengjie/miniconda3/lib/python3.9/site-packages/xoscar/backends/context.py:203: in send
    result = await self._wait(future, actor_ref.address, send_message)  # type: ignore
/home/lichengjie/miniconda3/lib/python3.9/site-packages/xoscar/backends/context.py:92: in _wait
    return await future
/home/lichengjie/miniconda3/lib/python3.9/site-packages/xoscar/backends/context.py:83: in _wait
    await asyncio.shield(future)
xoscar/serialization/core.pyx:913: in xoscar.serialization.core.deserialize
    ???
xoscar/serialization/core.pyx:811: in xoscar.serialization.core._deserial_single
    ???
xoscar/serialization/core.pyx:106: in xoscar.serialization.core.Serializer.deserial
    ???
/home/lichengjie/miniconda3/lib/python3.9/site-packages/xoscar/serialization/cuda.py:97: in deserial
    result = Serializable.device_deserialize(header, buffers)
/home/lichengjie/miniconda3/lib/python3.9/site-packages/cudf/core/abc.py:134: in device_deserialize
    frames = [
/home/lichengjie/miniconda3/lib/python3.9/site-packages/cudf/core/abc.py:135: in <listcomp>
    cudf.core.buffer.as_device_buffer_like(f) if c else memoryview(f)
/home/lichengjie/miniconda3/lib/python3.9/site-packages/cudf/core/buffer.py:125: in as_device_buffer_like
    return Buffer(obj)
/home/lichengjie/miniconda3/lib/python3.9/site-packages/cudf/core/buffer.py:188: in __init__
    ptr, size = get_ptr_and_size(np.asarray(buf).__array_interface__)
/home/lichengjie/miniconda3/lib/python3.9/site-packages/cudf/core/buffer.py:320: in get_ptr_and_size
    itemsize = cudf.dtype(array_interface["typestr"]).itemsize
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

arbitrary = '|S1'

    def dtype(arbitrary):
        """
        Return the cuDF-supported dtype corresponding to `arbitrary`.
    
        Parameters
        ----------
        arbitrary: dtype or scalar-like
    
        Returns
        -------
        dtype: the cuDF-supported dtype that best matches `arbitrary`
        """
        # first, try interpreting arbitrary as a NumPy dtype that we support:
        try:
            np_dtype = np.dtype(arbitrary)
            if np_dtype.kind in ("OU"):
                return np.dtype("object")
        except TypeError:
            pass
        else:
            if np_dtype not in cudf._lib.types.SUPPORTED_NUMPY_TO_LIBCUDF_TYPES:
>               raise TypeError(f"Unsupported type {np_dtype}")
E               TypeError: Unsupported type |S1

/home/lichengjie/miniconda3/lib/python3.9/site-packages/cudf/core/dtypes.py:51: TypeError