replace statsmodels with scipy stats

[martinkim0]

No description provided.

[martinkim0]

Tests are currently failing because pvalue has NaNs (before false_discovery_control is called). This wasn't being caught before for some reason (does statsmodels not check for NaNs?)

[justjhong]

@martinkim0 good catch, looks like statsmodels would just output nans. This happens when the input to chi2.cdf is 0 (which usually shouldn't happen - happens here since its barely training I assume), but we can still correctly handle this case by making the pval 1 for these examples.

[martinkim0]

Awesome, thanks for pushing the changes

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.14%. Comparing base (18b4432) to head (7256aca).

Additional details and impacted files

@@           Coverage Diff           @@
##             mrvi    #2762   +/-   ##
=======================================
  Coverage   89.14%   89.14%           
=======================================
  Files         167      167           
  Lines       14232    14232           
=======================================
  Hits        12687    12687           
  Misses       1545     1545           

Files	Coverage Δ
src/scvi/external/mrvi/_model.py	90.72% <100.00%> (ø)