* fix recordio split and task passes

* update for pre commit

* update

* update, still need to sync client wait for pass end.

* able to sync passes for task dispatching

* update to comment

* update

* fix yapf check

* why local pre-commit fails? version is the same

* fix race condition

* update

* fix race condition

* this still have duplicate problem in unit test

* update

* update

* update by comment

* update

Currently, WITH_GOLANG is a compiling flag, and not all Paddle binary
enable `WITH_GOLANG=ON`. So we cannot ensure that every paddle has
`master.so`. Make load that C libary lazily could let the user run
Paddle without `WITH_GOLANG=ON`.

* Refine unittest also

* Expose Net to Python

* Expose PlainNet to Python, make python can add_op, complete_add_op
* Provide a low level api to manipulate Net
* Unittest for Net::DebugString

modify the format

* add cross-entropy-op

* add infershape and compute

* implement Infershape and compute of onehotcrossentropy op

* a simplest SGD op

* Expose PlainNet to Python, make python can add_op, complete_add_op
* Provide a low level api to manipulate Net
* Unittest for Net::DebugString

2. Add crop layer unitest
3. Fix bugs

* Fc operator is a grouped operator, which combined by may internal
  operators.
* InferShape & Run a FC operator in Python.

* Follow review comments to seperate Tensor Numpy interactive methods in
  tensor.h.
* Simplify logic for `CastToPyBufferImpl`, make it as one struct and in
  details namespace.
* Remove `Scope` expose in Python, since it currently is useless.
* Remove some debug functions.

Basically following
http://pybind11.readthedocs.io/en/stable/advanced/pycpp/numpy.html

* Use buffer protocol to return a view of Tensor. It can be cast to
  numpy array in Python.
* Set a numpy array to a tensor.

Fix fetch record from master

* using etcd as fault tolerant training

* update

* workable version, ft not tested

* small fix

* update

* remove TODO