* add transforms.py and rename torchvision_reader.py to reader.py

* add datasets.py

* remove pytorch apis

* fix some small bugs

* remove torch and torchvision from requirements.txt

* modify core.CUDAPlace to fluid.CUDAPlace