* argsort op acceleration on GPU when the input size is equal to the length of the ‘axis’ dimension

* Optimize the error messages of paddle CUDA API, test=develop

* fix the error messages of paddle CUDA API, test=develop

* Refactoring PADDLE_ENFORCE_CUDA_SUCCESS, and apply to curand/cudnn/cublas/NCCL,test=develop

* remove build_ex_string,test=develop

* merge conflict,test=develop

* add backward gradient computation for op argsort test=developo

* use pre-commit test=develop

* add register op_data_type test=develop

* fix register bug in isfinite op test=develop

* rm int int64_t in pad2d gradKernel  test=develop

* Add ascending for argsort

* Refine api doc description.

* Refine descending description

* Add int32 logic to speedup when data is small size.

* Remove int32 opt as not support in python

* Improve argsort performance.

- Give 200000 data to compute argsort on v100,
can speed up ~190x
before opt cost: 0.53s
after opt cost:0.0027s

- Add fp16 support

* Refine error message
* Refine code

test=develop
Signed-off-by: Nzhaoyuchen <zhaoyuchen01@baidu.com>

test=develop