use psycopg2 again for the non-async demo

on second thought for the regular demo, we should
likely use psycopg2 for now, as the overhead of spinning
up an event loop on every DB call for "async_fallback"
is likely going to have a negative impact.

_sqlalchemy/queries.py

@@ -24,8 +24,8 @@ def connect(ctx):
 
     if session_factory is None:
         engine = sa.create_engine(
-            f"postgresql+asyncpg://sqlalch_bench:edgedbbenchmark@"
-            f"{ctx.db_host}:{ctx.pg_port}/sqlalch_bench?async_fallback=true"
+            f"postgresql://sqlalch_bench:edgedbbenchmark@"
+            f"{ctx.db_host}:{ctx.pg_port}/sqlalch_bench"
         )
         session_factory = orm.sessionmaker(bind=engine, expire_on_commit=False)