feat: wire Typesense search into backend, fix sync performance data bug

sync_typesense.py:
- Fix query string replacement: was matching 'ST_X(l.geom) as lng' but
  QUERY_BASE uses 'l.longitude as lng' — KS2/KS4 lateral joins were
  silently dropped on every sync run

backend:
- Add typesense_url/typesense_api_key settings to config.py
- Add search_schools_typesense() to data_loader.py — queries Typesense
  'schools' alias, returns URNs in relevance order with typo tolerance;
  falls back to empty list if Typesense is unavailable
- /api/schools: replace pandas str.contains with Typesense search;
  results are filtered from the DataFrame and returned in relevance order;
  graceful fallback to substring match if Typesense is down

requirements.txt: add typesense==0.21.0, numpy==1.26.4

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

pipeline/scripts/sync_typesense.py

@@ -158,7 +158,7 @@ def sync(typesense_url: str, api_key: str):
         query = QUERY_BASE
         if select_extra:
             # Insert extra select columns before FROM
-            query = query.replace("ST_X(l.geom) as lng", "ST_X(l.geom) as lng,\n        " + ",\n        ".join(select_extra))
+            query = query.replace("l.longitude as lng", "l.longitude as lng,\n        " + ",\n        ".join(select_extra))
         query += joins
 
         cur.execute(query)