feat(11-03): cpvDivisions column — normalizer + backfill migration

Adds Tender.cpvDivisions String[] (@@index Gin) derived at ingestion time
via cpv-catalog.ts's divisionOf() — replaces exact-match cpvCodes
comparison with a typesafe, GIN-indexable hasSome target (FILTER-03,
Pitfall 2). Backfill migration 20260721150000_tender_cpv_divisions_backfill
applied locally: 1612/1671 rows populated across all observed divisions
(741 rows carry division '45' — Bauarbeiten); idempotent (second run:
UPDATE 0, ADD COLUMN IF NOT EXISTS / CREATE INDEX IF NOT EXISTS both skip
cleanly). Applied via docker exec psql + `prisma migrate resolve
--applied` + `prisma generate` against the local dev DB only — no
Docker deploy on the test server.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

apps/api/prisma/migrations/20260721150000_tender_cpv_divisions_backfill/migration.sql

@@ -0,0 +1,41 @@
+-- FILTER-03 (Pitfall 2): Tender.cpvCodes is stored in inconsistent formats
+-- in the live DB (verified sample: "45", "45000000", "45000000-7" all
+-- referring to the same CPV division "45" = Bauarbeiten; 638 distinct raw
+-- values). An exact-equality `cpvCodes has '45000000'` filter would miss
+-- rows stored in a different format. This migration:
+--   1. Adds the `cpvDivisions` String[] column (typesafe, GIN-indexable
+--      hasSome-match target — replaces raw-SQL per-element LIKE matching,
+--      RESEARCH Pattern 3 Option B / Don't Hand-Roll).
+--   2. Adds the matching GIN index (matches schema.prisma
+--      @@index([cpvDivisions], type: Gin)).
+--   3. Backfills the ~1671 pre-existing rows by deriving the distinct
+--      2-digit CPV division from every element of cpvCodes, mirroring
+--      cpv/cpv-catalog.ts's normalizeCpv()+divisionOf() logic exactly
+--      (strip non-digit characters, take the leading 2 digits).
+--
+-- Idempotent: the backfill UPDATE only touches rows whose cpvDivisions is
+-- still the column default ('{}') — safe to run more than once, and never
+-- overwrites a value already set by the (now-fixed) normalizer for newer
+-- ingests. No user input is interpolated (T-11-06/07) — pure derived-field
+-- UPDATE over stored data.
+
+-- AlterTable
+ALTER TABLE "Tender" ADD COLUMN IF NOT EXISTS "cpvDivisions" TEXT[] NOT NULL DEFAULT '{}';
+
+-- CreateIndex
+CREATE INDEX IF NOT EXISTS "Tender_cpvDivisions_idx" ON "Tender" USING GIN ("cpvDivisions");
+
+-- Backfill: distinct leading-2-digit division per row, derived from every
+-- element of cpvCodes (implicit LATERAL unnest — a set-returning function
+-- in FROM may refer to a preceding FROM item without the LATERAL keyword).
+UPDATE "Tender" t
+SET "cpvDivisions" = sub.divisions
+FROM (
+  SELECT "Tender".id AS id,
+         array_agg(DISTINCT left(regexp_replace(code, '\D', '', 'g'), 2)) AS divisions
+  FROM "Tender", unnest("Tender"."cpvCodes") AS code
+  WHERE length(regexp_replace(code, '\D', '', 'g')) > 0
+  GROUP BY "Tender".id
+) AS sub
+WHERE t.id = sub.id
+  AND (t."cpvDivisions" IS NULL OR t."cpvDivisions" = '{}');

apps/api/prisma/schema.prisma

@@ -269,6 +269,7 @@ model Tender {
   title          String
   buyerName      String?
   cpvCodes       String[]  @default([])
+  cpvDivisions   String[]  @default([]) // FILTER-03: normalized 2-digit CPV divisions (hasSome-filterable, Pitfall 2)
   region         String?
   plz            String?
   bundesland     String?
@@ -287,6 +288,7 @@ model Tender {
   @@index([deadlineAt])
   @@index([publishedAt])
   @@index([bundesland]) // FILTER-02: post-backfill Bundesland-Filter-Performance
+  @@index([cpvDivisions], type: Gin) // FILTER-03: post-backfill CPV-Divisions-Filter-Performance (hasSome)
   // Deliberately NO tenant column and NO tenant index — this is global data (D-03)
 }