ELEX-3394 Part 1: distinguish between unexpected and non-predictive units

[dmnapolitano]

Description

Hi! The changes in this PR:

• refactor the way we determine reporting units, non-reporting units, and unexpected units so that _get_unexpected_units() is only called once rather than three times (which was once per each of reporting, non-reporting, and unexpected);
• introduce the predictive indicator in the dataframe of unexpected units, where True means the unit is genuinely unexpected but predictive, and False means otherwise (currently above/below the turnout factor thresholds);
• log a dictionary of non-predictive units that looks like {postal_code : [list of geographic_unit_fips]};
• provide unit tests for these

🎉

Jira Ticket

ELEX-3394 although this only resolves the "Thoughts" section of the ticket 😄 (Hence "Part 1")

Test Steps

tox or here's a CLI command that returns one non-predictive unit:

elexmodel 2022-11-08_USA_G --estimands=margin --features=baseline_normalized_margin --office_id=G_county --geographic_unit_type=county --pi_method bootstrap --national_summary --save_output results

[dmnapolitano]

Hi @lennybronner I just pushed the following changes:

• For units that are expected, regardless of whether or not they're reporting, their unit_category is now "expected" (instead of "reporting" or "non-reporting");
• The _get_units_without_baseline() method has been renamed to _get_units_with_baseline_of_zero() and now units with baseline_turnout == 0 are considered "non-modeled" units as opposed to "unexpected" units;

🎉