use fast executor as default (#17044)

test=develop

paddle/fluid/framework/details/execution_strategy.h

@@ -29,7 +29,7 @@ struct ExecutionStrategy {
   // this will loss 15%+ performance.
   // Please be aware about this parameters.
   size_t num_iteration_per_drop_scope_{1};
-  ExecutorType type_{kDefault};
+  ExecutorType type_{kExperimental};
   bool dry_run_{false};
   size_t num_iteration_per_run_{1};  // only use with async_ssa_graph_executor
                                      // and pyreader with data queue