Add a sequence slice layer.

Move pybind from package paddle/framework into paddle/pybind.

Identity operator and its gradient

Feature/extract op info into op info.cc

enable layer group to output a sequence inside it during generation.

Correctly handle width and height for DataLayer and ScatterAgentLayer.

Gather_op with python op passed

fix code style

Implement GPU kernel for cross entropy operator.

Rename `Net::AddOp` to `Net::AppendOp`

refine random related operators

* fix remote large update core

* wip

* working version

* fix style check

* fix style check

* update style check

Allow boot_bias for recurrent group to be static

Fix rowwise add