* polish r operator docs

* fix on naming convention

* Init commit

* Make executor use ProgramDescBind

* Change Attribute from BlockDesc to BlockDescBind

* Since we will get the program desc in RNN, just BlockDesc is not
  enough.

* Add DeviceContext to Executor API

* Rewrite RNN

* Pass Python

* AddBiasOp does not care num_flatten_dims

* Stash

* Fix MacOS Compile

* Pass RNN forward

* add python test

* refactor test

* Make compile pass

* add gradopmaker

* First draft done

* Polish code

* add grad op maker and grad infershape

* Polish code

* Fix backward.cc bug

* Fix infershape

* Rename function

* add backward test

* simplify recurrent test

* Update

* Pass unittest

* Add comments & refine test

* Add comments

* refactor test

* Complete Unittest

* fix StepScopes enforce

* Remove unused unittest

* no type error

* Update

* Make RNN Pass unittest

* Remove template parameter for Tensor methods

* Also check the type is correct when data()
* Simplize holder_

* Fix accuracy_op

* Register Code

* init

* unify CopyFrom interface

* fix gpu build error

* fix bug in tensor_py.h

* refine code comments and add TODO list

* fix conflicts in FeedOp and FetchOp

InferShape in Operator should be performed in OperatorBase::Run.

* cond_op, recurrent_op and mnist might be changed in following PR

* remove alias

Fix cycle dependencies, Fix #3583.

* Also simplify pybind implementation by using OperatorBase as holder
  type.

* make stepnet member

* add pybind support

* fix Inputs Outputs

* remove unique_ptr

* finish forward debug

* move namespace rnn to a directory

* move net_op to operators

* Make interface of Operator to `const Scope&`

* the tensor only contains shape and does not hold memory when inferring shape.