Full support for Python anndata >= 0.13

[rcannood]

Follow-up to #499 and PR #510.

PR #510 made py_to_r() and ReticulateAnnData work with the anndata.AnnData class name used by Python anndata >= 0.13, but anndataR is still not fully compatible with it -- it currently emits a "not fully supported" warning and the TEMP(anndata<0.13.0) pins from PR #481 are still in place.

Remaining work:

• track down and fix the remaining incompatibilities with anndata >= 0.13 (see the warning added in PR Class rename with anndata >= 0.13 and add warning #510; the vignette failure in usage_python vignette fails with mudata >= 0.4 #511 may be related)
• drop the TEMP(anndata<0.13.0) pins in tests/testthat/helper-skip_if_no_anndata_py.R, vignettes/usage_python.Rmd and .github/workflows/bench-base.yaml
• remove the warning in ReticulateAnnData once everything passes

[vivekbhr]

Dear maintainers

I think my issue is also relevant here. I find that the anndataR is not compatible with the .h5ad files written with anndata version 0.13 due to it not being able to handle nullable_string_array.

In anndata the .obs and .var indicies are nullable_string_array class by default. I had to fix this issue by modifying my anndata object to make it compatible with anndataR.

in R:

ad = anndataR::read_h5ad("input.h5ad")
Warning messages:
1: Error reading element obs of type <dataframe>
ℹ No function for reading H5AD encoding <nullable-string-array> for element
"obs/_index" 
2: Error reading element keys for obs of type <dataframe>
ℹ No function for reading H5AD encoding <nullable-string-array> for element
"obs/_index" 
3: Error reading element keys for var of type <dataframe>
ℹ No function for reading H5AD encoding <nullable-string-array> for element
"var/_index" 
4: Error reading element var of type <dataframe>
ℹ No function for reading H5AD encoding <nullable-string-array> for element
"var/_index" 

fixed in python:

ad.settings.allow_write_nullable_strings = False

adata = ad.read_h5ad("input.h5ad")
# modify index type
adata.obs.index = pd.Index(adata.obs.index.to_numpy(dtype=str))
adata.var.index = pd.Index(adata.var.index.to_numpy(dtype=str))

# Strip nullable string and categorical types from obs/var columns
for df in [adata.obs, adata.var]:
  for col in df.columns:
    if isinstance(df[col].dtype, (pd.CategoricalDtype, pd.ArrowDtype)) or "string" in str(df[col].dtype).lower():
      df[col] = df[col].astype(str).to_numpy(dtype=str)

adata.write_h5ad("output_fixed.h5ad")