fix(group): apply config when require_array returns an existing array

[d-v-b]

Part of the kwargs audit, here's a parameter (config) not being propagated correctly through the chain of function calls. this PR fixes that.

🤖 AI text below 🤖

Group.require_array (sync and async) forwarded **kwargs, including config, only to create_array in the branch that creates a missing array. When the array already existed, config was silently dropped:

g = zarr.group(zarr.storage.MemoryStore())
g.create_array("a", shape=(4,), dtype="uint8")
r = g.require_array("a", shape=(4,), dtype="uint8", config={"write_empty_chunks": True})
r.config.write_empty_chunks  # False before this PR, True after

This is the same bug as #4408 (open_array), found in the same **kwargs audit.

Changes

• config is a named parameter of Group.require_array and AsyncGroup.require_array, passed explicitly to both branches. Reading it from **kwargs is what let the existing-array branch drop it.
• New array: config goes to create_array.
• Existing array: after the shape and dtype checks, the array is returned through .with_config(config). The existing array comes from getitem, which serves member lookup in general (including consolidated metadata) and takes no runtime config, so adding one there would be out of place. with_config reuses the metadata and only rebuilds the codec pipeline.
• One parser for both branches: the existing-array branch runs config through parse_array_config before with_config, so an unknown key raises the same TypeError("Unknown array config keys: [...]") whether the array exists or is created, and missing keys come from the global configuration in both cases.

Behaviour change: an invalid config given for an existing array used to be ignored, and now raises (TypeError for an unknown key, ValueError for an invalid value).

Tests

• test_require_array_config: zarr format 2/3 × sync/async × existing/new array × config as None, a partial dict, a multi-key dict, or an ArrayConfig.
• test_require_array_unknown_config_key and test_require_array_invalid_config_value: one test per error case, each run against an existing and a newly created array.

14 of the new cases fail on main. The full test suite passes locally, and so does mypy.

Not covered

The other kwargs are still ignored for an existing array without a warning: stored-metadata arguments such as fill_value, chunks, attributes and overwrite, and unknown keys (which raise only on the create branch). How those should behave is an open question, and this PR doesn't answer it.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.45%. Comparing base (9e631c6) to head (9a025b2).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4413   +/-   ##
=======================================
  Coverage   94.44%   94.45%           
=======================================
  Files          93       93           
  Lines       13221    13224    +3     
=======================================
+ Hits        12487    12491    +4     
+ Misses        734      733    -1     

Files with missing lines	Coverage Δ
src/zarr/core/group.py	95.55% <100.00%> (+0.12%)	⬆️

🚀 New features to boost your workflow:

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