* update topi/cuda schedules to use target.max_num_threads

* allow num_thread to be larger than cuda.max_num_threads

* remove get_max_num_threads and make it inline