cuda: drop optimization

Returning early is confusing since the condition would rather be whether
_starpu_get_multi_worker_task returned non-0. The workers loop
terminates very quickly otherwise anyway, so avoid the confusion.

src/drivers/cuda/driver_cuda.c

@@ -950,9 +950,6 @@ int _starpu_cuda_driver_run_once(struct _starpu_worker_set *worker_set)
 #ifdef STARPU_SIMGRID
 	if (!res)
 		starpu_pthread_wait_wait(&worker0->wait);
-#else
-	if (!res)
-		return 0;
 #endif
 
 	for (i = 0; i < (int) worker_set->nworkers; i++)