cellxgene-schema CLI must validate schema_version

[brianraymor]

Context

See single-cell-four.

As part of automated migration, the schema transitioned from curators annotating schema_version to CELLxGENE Discover annotating schema_version. For 3.1, schema_version is temporarily overwritten as part of the transition:

When a dataset is uploaded, CELLxGENE Discover MUST automatically add the schema_version key and its value to uns. If schema_version is already defined, then its value MUST be overwritten.

In 4.0, curators must not annotate this field. An error should be thrown blocking validation if it is provided, rather than the overwriting behavior it currently uses.

[nayib-jose-gloria]

@jahilton Ready for testing! For all testing, you can pull the latest release candidate version of 4.0.0 by running:
pip install git+https://github.com/chanzuckerberg/single-cell-curation/@main#subdirectory=cellxgene_schema_cli

[jahilton]

Unexpected behavior: the file will pass validation if uns['schema_version'] = None
Expected to fail because 'schema_version' is present in uns.

Sample adata was set by adata.uns['schema_version'] = None & writing to .h5ad

[nayib-jose-gloria]

@jahilton mind double checking that the uns['schema_version'] = None actually was written to the h5ad? I tried to recreate locally, and it seems like anndata doesn't actually save dict keys with None values to the output h5ad when you use the '.write' function (someone else reported this issue here)

[jahilton]

You are correct. It drops when saving so the tested .h5ad didn't actually have uns.schema_version.
In that case, all clear!