fix(sharding): flatten any multi-dim value for coordinate selections in partial writes

[d-v-b]

Summary

Fix a dumb singleton dimension bug that breaks oindex with mixed np.array / integer components with the sharding codec.

Closes #4315

Claude wrote this, see the original PR here: d-v-b#320

Author attestation

• I am a human, these are my changes, and I have reviewed and understood every change and can explain why each is correct.

TODO

• Add unit tests and/or doctests in docstrings
• Add docstrings and API docs for any new/modified user-facing classes and functions
• New/modified features documented in docs/user-guide/*.md
• Changes documented as a new file in changes/
• GitHub Actions have all passed
• Test coverage is 100% (Codecov passes)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.23%. Comparing base (9c29a0d) to head (23a0003).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4316      +/-   ##
==========================================
+ Coverage   94.21%   94.23%   +0.01%     
==========================================
  Files          92       92              
  Lines       12871    12880       +9     
==========================================
+ Hits        12127    12137      +10     
+ Misses        744      743       -1     

Files with missing lines	Coverage Δ
src/zarr/codecs/sharding.py	96.25% <100.00%> (+0.02%)	⬆️
src/zarr/testing/strategies.py	95.78% <100.00%> (+0.35%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[d-v-b]

🤖 AI text below 🤖

Two follow-up commits, addressing the review finding on mask assignment and the test gap it exposed.

49f8872 narrows the flatten

The first version ravelled any multi-dimensional value once the shard-level indexer was a CoordinateIndexer. That is too broad: the shard-level selection cannot tell orthogonal from mask indexing, so a mask write with a (2, 2) value for four selected elements, or an orthogonal write with a spurious trailing axis like (3, 2, 1), was silently accepted on a sharded array while raising ValueError on an unsharded one (and in numpy).

The value shape can tell them apart. An np.ix_ selection has an N-D sel_shape, and the caller's value is that shape with only length-1 axes removed, which is exactly where np.ix_ puts integer indices. The helper now ravels only when value.shape is sel_shape minus zero or more unit axes (_drops_only_unit_axes). A mask arrives with a 1-D sel_shape, so a 2-D value no longer qualifies and fails downstream the same way it does unsharded; an extra axis fails to qualify for the same reason. test_sharding_set_rejects_value_with_wrong_rank covers both on both pipelines and fails against the previous commit.

aaf30c8 makes the property tests able to find this

The question of how this got past Hypothesis had a plain answer: the case was unreachable.

• test_oindex, test_mask_indexing and test_block_indexing skipped their set half on sharded arrays with assume(zarray.shards is None), added in Add more setitem property tests #2825 when bug with setitem with oindex and sharding #2834 was filed and never lifted. test_vindex had its set half commented out.
• orthogonal_indices wrapped every bare integer as a one-element array, so zarr never received an integer index and the dropped-axis path in OrthogonalIndexer was unreachable. basic_indices(min_dims=1) never yields an integer either, so that branch was dead.
• arrays() only drew a shard shape when every axis had a chunk strictly between 1 and the axis length. With the v3 and regular-grid draws on top, 2 of 500 test_oindex examples were sharded.

The commit lifts the skips, draws integers explicitly (the numpy indexer gets a 0-d index so its result drops the axis as oindex does), enables the vindex write behind a duplicate-point filter, and lets any chunk that fits the array be sharded (33 of 500 now, with an event() so --hypothesis-show-statistics shows the split). With these changes test_oindex fails against the code before this PR with the mixed-integer shape mismatch at the default 300 examples, and passes with it. The full property module and the stateful store and hierarchy machines pass under the default and nightly profiles.

The mask leniency is still outside what the property tests express, since they only draw correctly shaped values; the unit error test carries that.

[d-v-b]

it's going in! this unblocks a longstanding hole in our array indexing coverage for sharded arrays.