Check if zarr store supports consolidated metadata

[TomNicholas]

Needed to support the addition of .supports_consolidated_metadata in zarr-python upstream: zarr-developers/zarr-python#3119.

• 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

[TomNicholas]

TODO: I don't think I even want to try and define this class if zarr-python v3 is not installed

[jhamman]

Suggested change

	if not store.supports_consolidated_metadata:
	if not getattr(store, "supports_consolidated_metadata", True):

I think this will be needed for Zarr >3.0.0,<3.1.0

[TomNicholas]

Right - I had something like that then removed it 🤦

[TomNicholas]

added in a47b2e2

[paraseba]

Thank you @maxrjones . Long term, I'd love to see the responsibility for managing the readonly flag removed from the abstract Store hierarchy. It feels this is a concern that is beyond what a Store should know about. It would help us keeping the Stores abstract and focused. Readonly-ness sounds like a session concern, independent of the details of how to write zarr objects to different media.

[TomNicholas]

Thanks for the discussion everyone - I want to move any further discussion of this over to zarr-developers/zarr-python#3186, where I've raised @paraseba's suggestion as an issue.

For now I think I have what I need to be able to finish this PR.

[TomNicholas]

So for this PR obviously there are failures here, but:

• Upstream includes unreleased zarr data types changes, which will break a lot of things until @ianhi 's PR Updates for Zarr 3 Dtypes #10456 goes in - these are not needed for compatibility with zarr 3.0.9
• Most of the failures are easy to fix (things like regexes getting differently-worded error messages than they expected) EDIT: Apparently @dcherian already fixed this one in main in Update error message regex for latest Zarr #10485, tracked by remove xfail on test_to_zarr_zip_store in test_backends_datatree #10486
• I think the AttributeError: Can't get local object 'TestZarrNoConsolidatedMetadataSupport.create_zarr_target.<locals>.NoConsolidatedMetadataSupportStore' were caused by me hiding the definition of my test store in the test itself, and don't indicate a real bug EDIT: Defining it inside caused it to be unpickleable, which breaks some of the tests
• There are still Store is not read-only but mode is 'r', for tests where we use ZipStore, because IIUC that does not implement .with_read_only. As this is a breaking change the only way to work around it would be to change xarray's tests to use a read-only zip store, and that should technically be done in a separate PR to this one. EDIT: See How to roundtrip to a ZipStore? zarr-developers/zarr-python#3194 and the temporary workaround remove xfail on test_to_zarr_zip_store in test_backends_datatree #10486
• There is a failure where reading a non-consolidated store does not warn when it's expected to warn - I think that does indicate a logic bug in this PR for me to fix EDIT: addressed in 101b280, the test was expecting a warning that doesn't makes sense to be thrown for a store which does not support consolidated metadata

But that's it - then xarray should work with zarr 3.0.9 (and icechunk).

[TomNicholas]

• There is also an significant upstream performance regression caught by tests here that I am trying to ignore in Downgrade zarr requests performance expectations for zarr 3.0.9 #10491