feat(api): group GIAS school types and religions for parents

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/superpowers/specs/2026-10-02-school-type-groups-and-faith-filter-design.md

@@ -103,8 +103,8 @@ Christian schools that are not tied to one church.
 Grouping lives in the backend, not dbt. The API already translates GIAS codes
 to names when it loads the marts (`backend/data_loader.py:
 translate_gias_code_columns`), and every filter is applied to that DataFrame.
-Grouping at the same point needs no mart change, so there is no Airflow run
-between merge and staging showing it.
+Grouping there needs no mart change, so there is no Airflow run between merge
+and staging showing it.
 
 ### `backend/school_groups.py` (new)
 
@@ -112,25 +112,22 @@ between merge and staging showing it.
 - `UNOFFERED_TYPE_CODES`: the not-offered codes, so that "every code is
   accounted for" is testable.
 - `FAITH_GROUPS`: ordered `(key, label, frozenset[int])` per faith group.
-- `type_group_for(code) -> str | None` and
-  `faith_groups_for(code) -> tuple[str, ...]` (a missing code gives
+- `type_group_for(name) -> str | None` and
+  `faith_groups_for(name) -> tuple[str, ...]` (a missing or blank name gives
   `("none",)`).
 
 No dependence on the generated GIAS dictionaries beyond their codes, so the
 backend/pipeline dictionary parity test is untouched.
 
-### `backend/data_loader.py`
+### Grouping by name, at filter time
 
-In `translate_gias_code_columns`, before the code columns are replaced by
-names, add two columns from the codes:
-
-- `school_type_group`: `type_group_for(school_type_code)`, or None.
-- `faith_groups`: `faith_groups_for(religious_character_code)`, a tuple.
-
-The fallback query that reads name columns from older marts
-(`_MAIN_QUERY_NO_EXTRA_COLS` and its replacements) has no codes. There, both
-columns are derived from the names by reverse lookup through `SCHOOL_TYPE` and
-`RELIGIOUS_CHARACTER`.
+The groups are defined over codes but looked up by the translated name
+(`type_group_for(name)`, `faith_groups_for(name)`), when `/api/schools` and
+`/api/filters` filter. The DataFrame those endpoints read carries names only:
+`translate_gias_code_columns` replaces the codes at load, the legacy-name
+mart fallback never had codes, and the API test fixtures are written in
+names. The names come from the same dictionaries, so the lookup is exact.
+`data_loader.py` is unchanged.
 
 ### `/api/schools`