feat: renumber duplicate keys only on relevant course fields

[nm3224]

Primary Keys for course records now also include: section_id.
BUT: if section_id has >75% null/missing values, exclude it from primary keys and proceed without it.

PDP and Edvise schema duplicate handling now share a clear rule: suffix course_number only when the duplicate-key group disagrees on course classification/type, course name, credits attempted or earned, or grade (relevant fields we care about). Otherwise - collapse to one row per key (arbitrary dropping). The kept row is chosen deterministically as the first index in the group.

How this changes our student-term features:
For scalar ("course_id", catalog) in sum_val_equal_cols_by_group, course_id values that are renumbered variants of the catalog id (e.g. ENG101-1 for ENG101) are counted toward num_courses_course_id_*, so key-course counts can reflect multiple rows in a term after dedupe. This captured important information around students re-taking courses which was otherwise dropped.

• Q: Should this not be the case for lab/lecture combos or courses where the course-name is tangibly different, since those aren't really courses re-taken?
• • Note: Section_ID should make the grain more unique so that re-taken courses don't need to be suffixed if their section ID is different. If section ID is the same, then suffixing would still occur- not sure if this is the behavior we want since in that case the duplicate may be a "true" one / an error.

New unit tests added to ensure things are working as expected.