BUG: Consistent chunking with tensorstore across writes

[thewtex]

Different regions of an array could previously try to specify different
chunk dimensions. Use only the chunking of the region when creating the
dataset.

Fixes #161

[Copilot]

Pull Request Overview

This PR ensures consistent chunk metadata when using TensorStore, refactors TensorStore spec creation, and adds comprehensive tests for chunk consistency and zero-size validation.

• Added three pytest cases to verify TensorStore chunk shape consistency, sharding behavior, and zero‐chunk handling.
• Refactored _write_with_tensorstore to only include chunk metadata on initial dataset creation, build a minimal spec for existing stores, and retry writes on dimension/shape errors.
• Updated sharding logic in to_ngff_zarr and related helpers to standardize shard/chunk parameters and clean up sharding kwargs.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
py/test/test_to_ngff_zarr_tensorstore.py	New tests for TensorStore chunk shape consistency and zero‐chunk validation
py/ngff_zarr/to_ngff_zarr.py	Refactored TensorStore writing spec, exception handling, shard/chunk config

Comments suppressed due to low confidence (1)

py/test/test_to_ngff_zarr_tensorstore.py:149

• The skip reason mentions 3.0.0b1 but the version check is against 3.0.8. Update the reason to accurately reflect the required minimum version.

    zarr_version < version.parse("3.0.8"), reason="zarr version < 3.0.0b1"