Assess the quality of the statistical tests in tests/stats.py

[axch]

• Unit tests wouldn't hurt
• Check discrimination strength for some known test distributions
• Would be great to check that they catch errors that correspond to easy code typos: sign reversals, omitting terms, accidental doublings or halvings
  • This seems difficult to do generically without parsing the Python codes and automatically fuzzing them
  • Might also be good to check against extra roundings to, say, 5 decimal places, as a proxy for "subtle numerical errors"