groceries: relevance-ranked search (plain/basic products first)

- search.ts: score on the core name (brand/size stripped): exact or query+base suffix
  (kipdij+filet), head-final compounds, prefix, inside; penalties per extra word and per
  flavour/processing word (search-words.ts, editable); bonuses for store brands, the
  query's dominant category, store count, favourites; trigram tiebreak
- groceries.synonyms table (seeded: kipdij/kippendij, pasta → spaghetti, …) expands queries
- list rows show all pack sizes of a group; group picker uses the same ranking
- scripts/groceries-search-eval.ts prints top results + latency; preflight now 8 tables

scripts/groceries-search-eval.ts

@@ -0,0 +1,64 @@
+// Print the top results for some grocery searches with the relevance ranking (search.ts).
+// Read-only. Usage: DB_URL=… bun scripts/groceries-search-eval.ts [query …]
+// Uses the synonyms table, or the seed in the migration when the table doesn't exist yet.
+import { readdirSync, readFileSync } from 'node:fs';
+import postgres from 'postgres';
+import {
+	expand,
+	queryWords,
+	rankedGroups,
+	searchConds,
+	synonymMap,
+	type Synonyms
+} from '../src/lib/server/groceries/search';
+
+const sql = postgres(process.env.DB_URL!, { max: 2 });
+const queries = process.argv.slice(2).length
+	? process.argv.slice(2)
+	: ['kipdij', 'melk', 'rijst', 'tofu', 'kaas', 'eieren', 'pasta'];
+
+let synonyms: Synonyms;
+try {
+	synonyms = synonymMap(await sql`select term, synonym from groceries.synonyms`);
+} catch {
+	const dir = readdirSync('drizzle').find((d) => d.endsWith('_groceries_search_synonyms'))!;
+	const seed = readFileSync(`drizzle/${dir}/migration.sql`, 'utf8');
+	const rows = [...seed.matchAll(/\('([a-z0-9 -]+)', '([a-z0-9 -]+)'\)/g)].map((m) => ({
+		term: m[1],
+		synonym: m[2]
+	}));
+	synonyms = synonymMap(rows);
+	console.log(`(synonyms table missing; using the ${rows.length} from the migration seed)`);
+}
+const cats = await sql<
+	{ id: number; parent_id: number | null }[]
+>`select id, parent_id from groceries.categories`;
+const ctx = { at: Date.now(), synonyms, parent: new Map(cats.map((c) => [c.id, c.parent_id])) };
+
+for (const q of queries) {
+	const words = expand(queryWords(q), synonyms);
+	const where = [
+		sql`g.merged_into_id is null`,
+		sql`(g.item_count > 0 or g.favorite)`,
+		...searchConds(sql, words)
+	].reduce((a, c) => sql`${a} and ${c}`);
+	const t = performance.now();
+	const ranked = await rankedGroups(sql, where, q, words, { ctx });
+	const ids = ranked.map((r) => r.id);
+	const ms = performance.now() - t;
+	const top = ids.slice(0, Number(process.env.TOP) || 5);
+	const rows = await sql<{ id: number; name: string; brand: string | null; store_count: number }[]>`
+		select id, name, brand, store_count from groceries.groups where id = any(${top}::int[])`;
+	console.log(`\n${q}: ${ids.length} candidates, ${ms.toFixed(0)} ms`);
+	for (const id of top) {
+		const r = rows.find((x) => x.id === id)!;
+		const sc = ranked.find((x) => x.id === id)!;
+		console.log(
+			`  ${r.name} | ${r.brand ?? '-'} | ${r.store_count} stores | #${r.id}` +
+				(process.env.DEBUG
+					? ` | score ${sc.score} text ${sc.text} pen ${sc.penalty} cat ${sc.categoryId}`
+					: '')
+		);
+	}
+}
+await sql.end();