Refine drop_out_op and batch_norm

* add memory switch mechanism in operator kernel switch

Rename output of softmax and activations

Add data type checking for io.save_inference_model()

* init data_transform

* complete DataTransform

* fix build error

* add data_transform_test

* add a register test for data_transform_fn

* use function to simulate registration macro

* add register macro

* update test

* clean code

* restore unrelated code

* update data transform test

* generate unique name for REGISTER_DATA_TRANSFORM_FN

* add const

* follow comment

* update KernelTypePair hash function

Polish `Scope::LocalVarNames`

Fix sendrecv ut

Add python wrapper for reduce_max and reduce_min

Cannot get var name recursive since they could be same.

Refine sgd-op

Fix transpose doc

unify the indentation of license

 Speed data reader for IMDB dataset.

Refine the activation type getting in the LSTM operator to speed.

Add "download mklml failed" into FAQ

refine pipe_reader

* "fix threadpool style"

* "remove header"

update alexnet training data

* "add data layout"

* "need kernel registry support"

* "fix data layout"

* "reorder include headers"

* "change enum to enum class"

* "fix CI"

add mkldnn fluid design doc