Use CUDA_ARCH_NAME=All in the paddle/scripts/docker/build.sh and remove 20 21(20) in cmake/cuda.cmake.

This changed was accidently reverted in a previous pull request.  This pull request adds back in the print_operators_doc to the docker image.

Discussed with Helin and Yi. Currently we have a binary "print_operators_doc" that is built that generates a JSON document. This JSON represents the API documentation of paddle, and will be used by PaddlePaddle.org to generate API documentation.

This issue is to request "print_operators_doc" to be added to the production paddlepaddle/paddle:latest image.

After discussion with Helin and Yi, this change adds "print_operators_doc" executable to the Paddle docker nightly image.  This docker image will be pulled by PaddlePaddle.org nightly job and will generate the operator documentation to be put on PaddlePaddle.org website.

* Add documentation of cross-compiling for iOS.

* Correst the typo in documentation of cross-compiling for raspberry pi.

* Set ANDROID_API to 21 when it is specified < 21 for arm64-v8a in build_android.sh.

* Check the input and print the usage in MergeModel.cpp.

* init add

* add topk op

* someupdate

* fix style check

* add test py file

* update top k cuda kernel

* follow comments

* remove debug print

* fix casting error

* fix casting error

* fix casting error

* fix rename bug...

* fix travis

Fix invalid paddle executable file path with pip install  

Otherwise the production image generation will fail due to deb not
found, or executable built from Go not found.

Fix C-API Compilation Error

* User must install Paddle python package before unittest.
* Or use docker to build Paddle

* Do not use `virtualenv` in Docker
* Split unittest into many target, make them parallel