Created by: qingqing01

1. Add conv2d_grad_grad_op

   • the formula derivation of conv2d_grad_grad also can refer https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/ops/nn_grad.py#L33
   • Now only support GPU, the CPU supporting will be added in next PR.

2. Extracte the cuDNN conv algo searching code in conv_cudnn_helper.h.

   • Now use it in conv2d_grad_grad
   • Will simply the searching code in conv2d and conv2d_grad in next PR.

3. Enhance and fix bug in unit testing of gradient_checker

4. Support to fetch empty variables

   • see the changes in paddle/fluid/operators/controlflow/fetch_op.cc and python/paddle/fluid/executor.py
   • If the fetched var is empty, will return None in Python.