fix: dask divide by zero error when empty dimensions are passed into partial autochunk spec

[charles-turner-1]

Description

I discovered this whilst playing with charles-turner-1/xrexpr#121 (expression rewriting).

TLDR; auto-chunking on zero length dimensions fails, because dask divides by zero. Same issue for Datasets & Variables:

import numpy as np

import xarray as xr

da = xr.DataArray(
    np.zeros((5, 1)),
    dims=("lat", "lon"),
).isel(lat=[])

# This is fine
da.chunk("auto")

# This raises
da.chunk({"lon": "auto"})

I've hidden the traceback in a detail block because it's pretty long.

Details

Traceback (most recent call last):
  File "/Users/u1166368/xarray/bug_repro/exercise_broken_auto.py", line 14, in <module>
    da.chunk({"lon": "auto"})
    ~~~~~~~~^^^^^^^^^^^^^^^^^
  File "/Users/u1166368/dask/.pixi/envs/default/lib/python3.14/site-packages/xarray/core/dataarray.py", line 1498, in chunk
    ds = self._to_temp_dataset().chunk(
        chunk_mapping,
    ...<5 lines>...
        from_array_kwargs=from_array_kwargs,
    )
  File "/Users/u1166368/dask/.pixi/envs/default/lib/python3.14/site-packages/xarray/core/dataset.py", line 2638, in chunk
    k: _maybe_chunk(
       ~~~~~~~~~~~~^
        k,
        ^^
    ...<7 lines>...
        from_array_kwargs=from_array_kwargs.copy(),
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/Users/u1166368/dask/.pixi/envs/default/lib/python3.14/site-packages/xarray/structure/chunks.py", line 105, in _maybe_chunk
    var = var.chunk(
        chunks,
        chunked_array_type=chunked_array_type,
        from_array_kwargs=from_array_kwargs,
    )
  File "/Users/u1166368/dask/.pixi/envs/default/lib/python3.14/site-packages/xarray/core/variable.py", line 2740, in chunk
    return super().chunk(
           ~~~~~~~~~~~~~^
        chunks=chunks,
        ^^^^^^^^^^^^^^
    ...<2 lines>...
        **chunks_kwargs,
        ^^^^^^^^^^^^^^^^
    )
    ^
  File "/Users/u1166368/dask/.pixi/envs/default/lib/python3.14/site-packages/xarray/namedarray/core.py", line 851, in chunk
    data_chunked = chunkmanager.from_array(ndata, chunks, **from_array_kwargs)  # type: ignore[arg-type]
  File "/Users/u1166368/dask/.pixi/envs/default/lib/python3.14/site-packages/xarray/namedarray/daskmanager.py", line 74, in from_array
    return da.from_array(
           ~~~~~~~~~~~~~^
        data,
        ^^^^^
        chunks,
        ^^^^^^^
        **kwargs,
        ^^^^^^^^^
    )  # type: ignore[no-untyped-call]
    ^
  File "/Users/u1166368/dask/dask/array/core.py", line 3669, in from_array
    chunks = normalize_chunks(
        chunks, x.shape, dtype=x.dtype, previous_chunks=previous_chunks
    )
  File "/Users/u1166368/dask/dask/array/core.py", line 3200, in normalize_chunks
    chunks = auto_chunks(chunks, shape, limit, dtype, previous_chunks)
  File "/Users/u1166368/dask/dask/array/core.py", line 3414, in auto_chunks
    size = (limit / dtype.itemsize / largest_block) ** (1 / len(autos))
            ~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
ZeroDivisionError: division by zero

I haven't opened an issue for this - I could see a straightforward fix so I just got cracking on it, but I did have a look and I couldn't find any extant issues mentioning it. Happy to open one for clarity!

I also stuck the test right below the chunking tests in the dataarray section - not totally sure it's the best place for it, open to suggestions!

• Not sure if this really counts as user facing?

Checklist

• Closes #xxxx
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst N/A

AI Disclosure

• This PR contains AI-generated content.
  • I have tested any AI-generated content in my PR. N/A
  • I take responsibility for any AI-generated content in my PR. N/A

I wrote all code in this myself, but had Claude Opus 5 review my changes a couple of times until I was satisfied its critiques weren't super helpful.

[welcome]

Thank you for opening this pull request! It may take us a few days to respond here, so thank you for being patient.
If you have questions, some answers may be found in our contributing guidelines.

[charles-turner-1]

I've had Claude Opus 5 take a look at my changes with an eye to finding any lurking bugs or completeness issues.

It's suggestion is to push these changes down into namedarray, which I'm currently investigating - looks like the current implementation doesn't generalise well - still raises if Dataset.chunk() or variable.chunk() is called in an analagous scenario.

I'll undraft this when I'm satisfied with my changes.

[charles-turner-1]

to revert autoformatter

[charles-turner-1]

remove redundant test