Fix schema init: use exec_driver_sql to avoid PL/pgSQL bind-param collision

SQLAlchemy's text() treats $1/$2 in PL/pgSQL blocks as bindparams and raises
InvalidRequestError. exec_driver_sql sends raw SQL directly to psycopg2,
bypassing that processing. Without this all tables are missing on startup.

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

backend/main.py

@@ -29,25 +29,21 @@ from scheduler import start_scheduler, shutdown_scheduler
 @asynccontextmanager
 async def lifespan(app: FastAPI):
     # Apply database schema (idempotent CREATE TABLE IF NOT EXISTS)
+    # Uses exec_driver_sql to bypass SQLAlchemy parameter parsing — schema.sql
+    # contains PL/pgSQL with $1/$2 syntax that text() misinterprets as bindparams.
     try:
         import os
-        from database import SyncSessionLocal
+        from database import sync_engine
         schema_path = os.path.join(os.path.dirname(__file__), 'schema.sql')
         if os.path.exists(schema_path):
-            db = SyncSessionLocal()
-            try:
-                with open(schema_path) as f:
-                    sql = f.read()
-                db.execute(text(sql))
-                db.commit()
-                logging.getLogger(__name__).info("Schema applied successfully")
-            except Exception as e:
-                logging.getLogger(__name__).warning(f"Schema init failed: {e}")
-                db.rollback()
-            finally:
-                db.close()
+            with open(schema_path) as f:
+                sql = f.read()
+            with sync_engine.connect() as conn:
+                conn.exec_driver_sql(sql)
+                conn.commit()
+            logging.getLogger(__name__).info("Schema applied successfully")
     except Exception as e:
-        logging.getLogger(__name__).warning(f"Schema load failed: {e}")
+        logging.getLogger(__name__).warning(f"Schema init failed: {e}")
 
     # Startup: clean up stale scrape batches
     try: