[ENH] explicitly exclude subjects with NA data before starting to run ModelArray.lm() or .gam()

[zhao-cy]

If a column in phenotypes is used in (full) formula, exclude subjects with NA.

This is to avoid full model and reduced model would include different list of subjects, which will cause inaccuracy in calculated changed.rsq. Suggested by Bart. See slack message with him on/around 1/31/22.

Currently, please avoid requesting calculation of changed.rsq (i.e., by requesting changed.rsq.term.index) for terms that have missing values.

[zhao-cy]

Temporary solution for now:

To users: ⚠️⚠️WARNING!!⚠️⚠️ When using ModelArray.gam(), if you want to request changed.rsq.term.index for a term that has missing values, before feeding phenotypes into ModelArray.gam(), you must exclude those observations (i.e., those rows in phenotypes) who have missing values in this term of interest from phenotypes. You should do the same for each term you'd like to request in changed.rsq.term.index, if that term has missing values. Without such exclusion, the full and reduced models would include different number of subjects, causing inaccuracy of calculation of delta.adj.rsq and partial.rsq.

Above warning will also show up in the ModelArray.gam() documentation after the branch is merged to main.