* Add softmax_with_cross_entropy and smooth_l1 in Python API.

* Fix doc format.

* dynamic send/recv selected rows

* update by comment

* fix by comment

* dist train support lr decay

* update by comment

* revert elementwise method creator

* delete comment

* Refine target_assign_op to unify the classification and regression targets assignment.

* Fix the unit testing.

* Fix conflicts.

move paddle/pybind/pybind.h to paddle/fluid/pybind/pybind.h, and cancel the test_parallel_op temporary