Fix scope of gather and broadcast, and code clean

fix reduce_op_handle_test

make for_parallel default True

"polish activation"

Add flush of program desc to update the proto information.

Feature/add reduce op handle

Refine gather and broadcast

Eager destruct local scope

VariableResponse support deserialize var into local scope

* Fix print op cpplint errors

* Remove commented code

fix split byref op mac build

Skip updating not trainable parameters in distribute transpiler

Remove intermediate output's gradient from inputs of grad_op.

check pserver file while writing dockerfile

* accelerate dropout

* accelerate dropout

* "fix the dropout test"

* "rerun ci"

* "fix ci"

* "rerun ci"

* "fix ci"

* "fix"

* "stage"

* disable

Auto-grown sparse table

`make install` does not remake whl package if only change .cc or .h

Fix the bug in executor.fetch_var

fix training parameter issue

Fix CPPLint errors in some framework files

update new_op readme