Extend return value for layer functions

Make users can access parameters of layers and their gradients.

* add update graph method for memory optimization transpiler to avoid rebuild graph everytime

* clean code

* reset var desc if hit cache

Adversarial sample

Add sequence erase op

Async GRPC sendrecv

fix ds2 issue

auto set openblas env

Add edit distance operator

* init data_type_transform

* split data_layout_transform

* tmp rm data_transform_test

* change device_data_transform to data_device_transform

* clean code

* clean code

Polish GetPlacesOp

Enhence shrink_rnn_memory_op.

feature/parallel_gpu

Calculating gradients for partial graph

* "init use all default devices"

* "fix init test"

* add lod tensor ToAbsOffset test

* add share lod to topk op and softmax op

Refine while grad op

Add grad_op_maker for sequence_pool

Add script to plot learning curve

Added backward.calc_gradient to backpropagate gradient from given targets to inputs.