bugfix: read fixed-length string arrays as bytes

[gszep]

There is a bug in pandas that prevents fixed-length string arrays np.dtype("S") from being read properly. For some reason the dataframe does the casting properly only if the array is set after dataframe creation.

x = np.array([b"123",b"e23423"], dtype="S")
df = pd.DataFrame({"a": x})
df.types # sees x as object
df = pd.DataFrame({})
df["a"] = x
df.dtypes # correct type

This PR is a workaround until the pandas bug is addressed

[gszep]

reported in pandas pandas-dev/pandas#51994

[codecov]

Codecov Report

Merging #982 (1794406) into main (dcee1e5) will increase coverage by 0.01%.
The diff coverage is 100.00%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #982      +/-   ##
==========================================
+ Coverage   84.15%   84.17%   +0.01%     
==========================================
  Files          34       34              
  Lines        4709     4713       +4     
==========================================
+ Hits         3963     3967       +4     
  Misses        746      746              

Impacted Files	Coverage Δ
anndata/_io/specs/methods.py	87.90% <100.00%> (+0.12%)	⬆️

[ivirshup]

Thanks for the PR!

If you use assert_equals in those failing tests, we'll probably get more useful error messages. I believe those tests predate that helper.

I would significantly prefer this being fixed upstream, and restricting usable versions of pandas, rather than having to do and then undo fixes + multiple releases here. Would you be up for tackling this upstream?

Also, is this issue new with pandas 2.0?