Clean code & add execution strategy

add boost to inference lib

Polish data_type.h

Delete prefetch_ctx_ after use.

Change default split minimize value to speed up transport data to pserver

* refine async update design doc

* update by comments

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into feature/polish_visit_data_type

Fix pe bug

* add dist transpiler unit test

* update by comment

* update by comment

* fix ci

* fix ci

* add manylinux docker image usage

* update by comment

* update by comment

refine EngineIOConverter, and use io_convert in test_trt_activation_op

Change cmake for build doc

* Fix a bug related to dispensable inputs and refine the inference unittest.

* Fix the use of dispensable inputs in reshape_op.

* Polish the enforce statements.

* Fix an English writing typo.

[Feature] NCCL2 distributed training