Update for cope with issue introduced in umap-learn 0.5.2

[lmcinnes]

As of umap-learn 0.5.2 an n_epochs value of 0 is now a valid value and None is used instead to indicate a value should be generated. To maintain forward compatability and backward compatibility with older umap-learn versions we instead make use of the epch setting criterion internal to umap-learn here directly. This should resolve issue #2026 and lmcinnes/umap#798

[codecov]

Codecov Report

Merging #2028 (a7d60e0) into master (1b74015) will increase coverage by 0.01%.
The diff coverage is 100.00%.

❗ Current head a7d60e0 differs from pull request most recent head 91aa733. Consider uploading reports for the commit 91aa733 to get more accurate results

@@            Coverage Diff             @@
##           master    #2028      +/-   ##
==========================================
+ Coverage   71.63%   71.64%   +0.01%     
==========================================
  Files          92       92              
  Lines       11248    11252       +4     
==========================================
+ Hits         8057     8062       +5     
+ Misses       3191     3190       -1     

Impacted Files	Coverage Δ
scanpy/tools/_ingest.py	68.33% <100.00%> (+0.12%)	⬆️
scanpy/tools/_score_genes.py	85.71% <100.00%> (+1.56%)	⬆️
scanpy/tools/_umap.py	74.24% <100.00%> (+0.39%)	⬆️

[Zethson]

Thanks!
@ivirshup umap-learn 0.52 was released 3 days ago. Maybe we should even push out a patch release? I expect that many people will run into this issue and be left confused.

[ivirshup]

Thanks for the fix!

I've just added a release note.