Fix sparse update memory error for distributed training

* Update tensor_util.h

* Update with moved TensorDesc

* Fix tensur_utils.cu

* Update

* Update

* Update

* Update

* Make tensor_util.cu a symbolic link

* dynamic send/recv selected rows

* update by comment

* fix by comment

* Polish sum_op support SelectedRows in_place

* test dist word2vec

* multiple trainers work

* Rename Tensor::CopyFrom to Tensor::Copy

* Fix CI

* Fix compile

There are mainly following fixes:

- take `DeviceContext` as the template parameter of math functors and OpKernel instead of `Place`
- remove `eigen_device` interface in base class  `DeviceContext`
- remove `GetEigenDevice` interface in `ExecutionContext` and base class `DeviceContext`
- remove unused `platform::EigenDeviceConverter`
- rename `REGISTER_OP_GPU_KERNEL` to `REGISTER_OP_CUDA_KERNEL`
- rename `USE_GPU_ONLY_OP` to `USE_CUDA_ONLY_OP`

* Add DataFeeder

A v2 API like data feeder for book demos.
We can feed data directly from reader.

* Fix CI

* Add an unittest for while/rnn op forward

* Add unittest for raw while op backward

* Fix CI

* Fix Proformance problem of enforce

* Fix missing `;` in code

* Fix CI

* "make global tensor function independently"

* "replace functor"

* "fix inline template error"

* "fix tensor array with CopyFrom"

* "fix other case use CopyFrom"

* "move the op interface hardly"

* "fix operators"

* "fix typo"

* "delete dynamic recurrent rnn and fix gru_unit in debugmode"

* "fix unique_ptr copy"

* "fix cuda copy"

* "fix namespace error"

* "removed nccl python test"

* "fix include error"

* "fix typo"

* fix copy util test

* Use stable_sort in lod_rank_table

It is easy to debug and test when use `stable_sort`and the time
complexity is not changed.

* Add LoDTensorArray

* Stash

* Better debug message for IsInitialized

* Stash

* Better debug message for IsInitialized

* Complete array read/write op unittests

* Add unittest, Gradient of array read/write

* Follow comments

* 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

* add sparse support for sum op

* typo fix

* fix gpu build error

* fix unittest error

* typo fix

* infer var type and shape in op_test

* follow comments

* fix build error

* bypass some unittests depend on NetOp

* support sparse output for lookup table grad op

* refine codes

* fix gpu build error

* fix lookup table grad gpu kernel

* fix ci

* fix ci

* fix ci

* fix bug in lookup_table_grad op

* fix bug in test_word2vec

* register double kernel for some operators

* set is_sparse=True in test_word2vec

* fix lookup table grad op CUDA kernel bug

* disable test_modified_huber_loss_op temporarily

* disable test_lstm_unit_op temporarily

* add sparse support for sum op

* typo fix

* fix gpu build error

* fix unittest error

* typo fix

* infer var type and shape in op_test

* follow comments

* fix build error

* bypass some unittests depend on NetOp

* No POD instance is forbidden in global scope.

See
https://google.github.io/styleguide/cppguide.html#Static_and_Global_Variables