fix: keep FlyBase gene nan when reading 10x mtx

[joncarse]

• Closes ValueError: cannot convert float NaN to integer #1259
• Tests included
• Release notes not necessary because:

FlyBase has a real gene named nan. read_10x_mtx reads genes.tsv / features.tsv with pd.read_csv, and pandas treats nan as missing by default, so that symbol disappeared. That showed up as calculate_qc_metrics blowing up after var_names.str.startswith(...) (#1259, also #1708), or as a silently wrong identifier.

This keeps lowercase nan as a gene name and still treats the rest of pandas’ NA tokens (NaN, NA, and so on) as missing. The default token list is copied in the reader so production code does not import pandas._libs; a test compares that copy to pandas so we notice if it drifts.

barcodes.tsv is unchanged.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.98%. Comparing base (a6f1a2d) to head (155cc30).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4329   +/-   ##
=======================================
  Coverage   81.98%   81.98%           
=======================================
  Files         134      134           
  Lines       13235    13237    +2     
=======================================
+ Hits        10851    10853    +2     
  Misses       2384     2384           

Flag	Coverage Δ
hatch-test.low-vers	79.07% <100.00%> (+<0.01%)	⬆️
hatch-test.pre	81.86% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/scanpy/io/_read.py	80.74% <100.00%> (+0.14%)	⬆️

[joncarse]

I doubt "nan" will ever appear as truly meaning a nullish value in any TSV file. If this is a concern we can technically filter out "nan" only for Drosophila analyses. But then again maybe someone will try to splice it into another organism :)

[joncarse]

Lines 137-141 specifically test regressions for cases such as #1259 , but might look out-of-context when reading these lines in the future. I can remove or extract to another unit test that specifically mentions #1259

[joncarse]

I wanted to protect against drift in future pandas versions but didn't want production code to import a private ("_") pandas member. Thought putting it as a test is a good tradeoff