Design organization_units database schema

Use an adjacency list (not a nested set or closure table) as specified — it is the simplest model for a hierarchy that changes infrequently and needs efficient parent→children traversal. The recursive CTE view is the right pattern for NHF's 3-level hierarchy (national → region → chapter). For circular reference prevention, a PostgreSQL trigger (BEFORE INSERT OR UPDATE) that walks the ancestor chain is the most reliable approach — the trigger calls a recursive function checking if the new parent_id is a descendant of the row being inserted. Keep the trigger simple: it only needs to walk up the tree (following parent_id) and abort if it ever reaches the current row's id.

NHF has up to 1,400 chapters — the hierarchy is wide, not deep (max 3 levels), so recursive traversal is cheap. Ensure the migration filename follows the existing project convention (timestamp prefix).

Write SQL-level tests (pg_tap or plain SQL assertions in a test migration): (1) insert a national → region → chapter chain and verify the CTE view returns correct depth values (0, 1, 2), (2) attempt to delete a region node that has chapter children and verify ON DELETE RESTRICT raises an error, (3) insert two chapters with the same name under the same parent and verify the unique constraint rejects the second insert, (4) insert a chapter with deleted_at set and verify the unique constraint does NOT block a new active chapter with the same name under the same parent, (5) verify that setting parent_id = id (self-reference) is rejected by the circular reference check. Also write a Dart integration test that calls OrganizationUnitRepository.getChildren(parentId) and verifies the returned list matches the seeded data.