Add read support for Zarr spatial and proj conventions

[emmanuelmathot]

Add read support for Zarr spatial and proj conventions

This is the second in a series of PRs splitting #883 as requested by maintainers. Depends on #899, #903, #904.

Changes:

• Add Convention.ZARR enum value to rioxarray.enum
• Create zarr.py module with ZarrConvention class implementing ConventionProtocol
• Register ZarrConvention in _core.py's _CONVENTION_MODULES for automatic convention detection
• Support proj:wkt2, proj:code, proj:projjson for CRS reading
• Support spatial:transform, spatial:dimensions for transform and dimension reading
• Convention declaration checking via zarr_conventions attribute

Architecture:

• Follows the protocol-based convention architecture established in Add Convention option & extract CF logic to cf.py #899/REF:convention: Reorganize convention modules & use rio accessor grid mapping methods #903
• ZarrConvention class implements read_crs(), read_transform(), read_spatial_dimensions()
• Write methods raise NotImplementedError (to be implemented in future PRs)

Reviewer feedback addressed:

• ✅ Both conventions are attempted when reading regardless of setting (priority changes only)
• ✅ Use CRS.from_user_input() for CRS parsing
• ✅ Use functional test format, not class format
• ✅ Imports at top of file
• ✅ Follows ConventionProtocol interface from REF:convention: Reorganize convention modules & use rio accessor grid mapping methods #903

---

• Closes Implement Zarr spatial and proj conventions support #883 (partial - second of series)
• Tests added (20 unit tests + 10 integration tests)
• Fully documented, including history.rst for all changes

Future PRs:

• PR 3: Writing Zarr CRS (write_crs(convention=Convention.Zarr))
• PR 4: Writing Zarr Transform (write_transform(convention=Convention.Zarr))

[emmanuelmathot]

Any updates on this one @snowman2? I dont know how to solve the failing tests.

[snowman2]

I need to first dig into the GeoZarr spec and want time to review the code before moving forward. Test failures that are unrelated won't block merging.

Possibly relevant:

• Zarr 3 (and geozarr?) OSGeo/gdal#11824

[snowman2]

https://xkcd.com/927/

[emmanuelmathot]

I understand. Few notes here:

• see this recent PR more relevant for our case: Zarr: implement read and write support for 'spatial' and 'proj' conventions OSGeo/gdal#13743
• This PR is not about GeoZarr (yet). THis is really about reading the underlying Zarr conventions. Please check that page for more context: https://explorer.eopf.copernicus.eu/datamodel/

[snowman2]

Thanks @emmanuelmathot 👍