Created by: zhouwei25

1. Speed up the compilation, remove the Useless intermediate process of paddle_fluid/paddle_fluid_origin/paddle_fluid_shared/memory/paddle_framwork , and change them to be library which does't cost time. As a result, the dependency relationship between targets has been adjusted, too.

2. Reduced some target_link_libraries of testing executables. Currently, there are unnecessary libs are linked into testing executables.

3. Eliminates compilation warnings caused by std=c++11.

4. Some message warnings of cmake have been removed to avoid disturbing users.

5. Add tensorrt information to version.txt of inference lib.

It has been checked by PR_Windows_CI_Inference http://ci.paddlepaddle.org/viewLog.html?tab=buildLog&buildTypeId=PaddleWindows_PrWindowsCiInference_2&buildId=270224&_focus=1443#_state=1442