cupy-xarray does not handle chunked dask arrays with NaNs

[yt87]

What happened?

This bug was reported here xarray-contrib/cupy-xarray#52. Repeating here on a request from @dcherian.
I added a print line in duck_array_ops.py, function as_shared_type:

    # Avoid calling array_type("cupy") repeatidely in the any check
    array_type_cupy = array_type("cupy")
    if any(isinstance(x, array_type_cupy) for x in scalars_or_arrays):
        import cupy as cp

        xp = cp
    elif xp is None:
        xp = get_array_namespace(scalars_or_arrays)
    print('=======', xp.__name__, scalars_or_arrays)

What did you expect to happen?

a is evaluated correctly, xp in as_shared_dtype is set to cupy, as expected.
b is also correct, it is a dask array with cupy.ndarray chunks. There is some black magic involved, since xp is set to numpy.
c is dask array with chunks of type numpy.ndarray. This is wrong, subsequent calls to as_numpy or compute fail with the traceback as shown in the bug report xarray-contrib/cupy-xarray#52.

Minimal Complete Verifiable Example

import numpy as np
import xarray as xr
import cupy_xarray

a = xr.DataArray([1, np.nan]).as_cupy().sum(min_count=1)
print('a =', a)
b = xr.DataArray([1, 2]).chunk(dim_0=2).as_cupy().sum(min_count=1)
print('b =', b)
c = xr.DataArray([1, np.nan]).chunk(dim_0=2).as_cupy().sum(min_count=1)
print('c =', c)

### MVCE confirmation

- [X] Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
- [X] Complete example — the example is self-contained, including all data and the text of any traceback.
- [X] Verifiable example — the example copy & pastes into an IPython prompt or [Binder notebook](https://mybinder.org/v2/gh/pydata/xarray/main?urlpath=lab/tree/doc/examples/blank_template.ipynb), returning the result.
- [X] New issue — a search of GitHub Issues suggests this is not a duplicate.
- [X] Recent environment — the issue occurs with the latest version of xarray and its dependencies.

### Relevant log output

```Python
======= cupy [array([0., 0.]), array([ 1., nan])]
======= cupy [nan, array(1.)]
a = <xarray.DataArray ()> Size: 8B
array(1.)
b = <xarray.DataArray 'asarray-a8e500e046402975cc20971c9b97fd57' ()> Size: 8B
dask.array<sum-aggregate, shape=(), dtype=int64, chunksize=(), chunktype=cupy.ndarray>
======= numpy [dask.array<zeros_like, shape=(2,), dtype=float64, chunksize=(2,), chunktype=cupy.ndarray>, dask.array<asarray, shape=(2,), dtype=float64, chunksize=(2,), chunktype=cupy.ndarray>]
======= numpy [nan, dask.array<sum-aggregate, shape=(), dtype=float64, chunksize=(), chunktype=cupy.ndarray>]
c = <xarray.DataArray 'asarray-a913867b040846d481995bbceaf8efdf' ()> Size: 8B
dask.array<where, shape=(), dtype=float64, chunksize=(), chunktype=numpy.ndarray>

Anything else we need to know?

No response

Environment

INSTALLED VERSIONS ------------------ commit: None python: 3.12.4 | packaged by conda-forge | (main, Jun 17 2024, 10:23:07) [GCC 12.3.0] python-bits: 64 OS: Linux OS-release: 6.9.3-3-MANJARO machine: x86_64 processor: byteorder: little LC_ALL: None LANG: en_US.UTF-8 LOCALE: ('en_US', 'UTF-8') libhdf5: 1.14.3 libnetcdf: 4.9.2

xarray: 2024.6.0
pandas: 2.2.2
numpy: 2.0.0
scipy: None
netCDF4: 1.7.1
pydap: None
h5netcdf: None
h5py: None
zarr: None
cftime: 1.6.4
nc_time_axis: None
iris: None
bottleneck: None
dask: 2024.6.2
distributed: 2024.6.2
matplotlib: None
cartopy: None
seaborn: None
numbagg: None
fsspec: 2024.6.0
cupy: 13.2.0
pint: None
sparse: None
flox: None
numpy_groupies: None
setuptools: 70.1.1
pip: 24.0
conda: None
pytest: None
mypy: None
IPython: 8.25.0
sphinx: None

[keewis]

I think this is is a duplicate of #7721. Can you confirm?

[yt87]

Yes, it is. I also think that without a solution, cupy-xarray is not a viable idea.

[keewis]

then let's close this and continue on #7721