Reassess use of "private" columns in `prep_samples_for_cohort_grouping` and `build_cohorts_from_sample_grouping`

[leehart]

Re: discussion #694 (comment)

[shauryam2807]

Hello @leehart @jonbrenas,
I just had PR #983 (for #806) approved, which addressed the sibling issue about filter_unassigned in the same code area. I'd like to take this one on too.

After reading the full PR #694 discussion, I see two approaches:

Option A (minimal): Normalize taxon_by like area_by/period_by — add df_samples["taxon"] = df_samples[taxon_by] in _prep_samples_for_cohort_grouping(), then remove the taxon_by param from _build_cohorts_from_sample_grouping(). This makes all three *_by params consistent.

Option B (thorough): Use private prefixed column names (e.g., _cohort_area, _cohort_period, _cohort_taxon) and add validation to reject user columns with the same prefix.

Option A is simpler and addresses the inconsistency. Option B is cleaner but more invasive.
Which direction would you prefer?

[jonbrenas]

Thanks @shauryam2807. I think Option A is the right approach, at least for now. We can revisit Option B later if we decide that it fits the needs better.

[shauryam2807]

Thanks @jonbrenas! I'll go with Option A — normalizing taxon_by to a standard "taxon" column, consistent with "area" and "period". Will open a PR shortly.

[shauryam2807]

Hello @jonbrenas
I've opened PR #997 implementing Option A as discussed. The changes normalize taxon_by to a standard "taxon" column in _prep_samples_for_cohort_grouping(), consistent with how area_by and period_by are handled. This also removes the taxon_by parameter from _build_cohorts_from_sample_grouping() and simplifies the label generation to a single code path.