* add reinforcement learning model test=develop

* align backward test=develop

* add gym in paddle_build.sh test=develop

* rm pip install in script test=develop

* refine paddle_build.sh test=develop

* fix sed error in macOS test=develop

* polish code test=develop