Created by: sfraczek

With this fix the slim int8 mkldnn quantization test (https://github.com/PaddlePaddle/Paddle/pull/17634) runs on multiple threads. This related issue: https://github.com/PaddlePaddle/Paddle/issues/17903.

@zhaify, @bingyanghuang - Do you think its ok that I renamed FLAGS_OMP_NUM_THREADS -> FLAGS_paddle_num_threads in CMake? There is such a flag defined already in Paddle so I wanted to reuse it since it has the same use. I didn't find OMP_NUM_THREADS flag.