GroupMetadata and ArrayV2Metadata should ignore extra keys for zarr v2 metadata

[TomAugspurger]

Zarr version

v3

Numcodecs version

na

Python Version

na

Operating System

na

Installation

na

Description

Some Zarr v2 implementations store additional fields at the top level of the metadata files, e.g. nczarr. zarr-python 3.x currently can't read these files because the deserialization methods pass all the keys to the GroupMetadata and ArrayV2Metadata constructors.

Steps to reproduce

In [10]: GroupMetadata.from_dict({"zarr_format": 2, "attributes": {"key": "value"}, "extra": "value"})
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[10], line 1
----> 1 GroupMetadata.from_dict({"zarr_format": 2, "attributes": {"key": "value"}, "extra": "value"})

File ~/gh/zarr-developers/zarr-python/src/zarr/core/group.py:119, in GroupMetadata.from_dict(cls, data)
    116 @classmethod
    117 def from_dict(cls, data: dict[str, Any]) -> GroupMetadata:
    118     assert data.pop("node_type", None) in ("group", None)
--> 119     return cls(**data)

TypeError: GroupMetadata.__init__() got an unexpected keyword argument 'extra'

Additional output

We can filter out unexpected keys before calling the constructors. We should decide what action to take when we encounter an unknown key (warning with the python logging module, a Python warning, and maybe some sort of config options to control the behavior).

For Zarr v3, I think the expectation is that extra keys are allowed, but they must be objects with a must_understand field.