fix PrePostNEGD input validation

[jpreszler]

This resolves issue #161

I tried the PrePostNEGD example with group being binary (current), boolean, or categorical with levels 'A' and 'B'. All three pass the _series_has_2_levels check but in the prediction after model fitting, errors result in the 'A'/'B' case because 'untreated' is given values of 0 (and 'treated' 1). Switching to _is_variable_dummy_coded moves the error to a data exception in the input validation, which makes more sense.

A more complicated fix would be to assign the two levels in lines 1059 and 1070 rather than 0's and 1's, but then we would need to know which is the 'treatment' level so probably best to leave those transformations to users.

I also fixed a bug that requires the group variable to be called 'group' and adjusted some re-used variable names whose names didn't match what was being stored.

[codecov]

Codecov Report

Merging #247 (526d8a8) into main (133b987) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##             main     #247   +/-   ##
=======================================
  Coverage   73.89%   73.89%           
=======================================
  Files          19       19           
  Lines        1149     1149           
=======================================
  Hits          849      849           
  Misses        300      300           

Files Changed	Coverage Δ
causalpy/pymc_experiments.py	66.75% <100.00%> (ø)

[jpreszler]

@drbenvincent also ready when you get a chance