add python-opencv in paddlepaddle/paddle:lastest images

* Enable assertions in CUDA.

* Fix PADDLE_ASSERT.

small clean up and document pointer ownership.

disable failed tests

Try to speed up parallel executor

Add image_resize_short and refine resize API

update memory profile url

* "fix deterministic"

* "fix ci"

* "fix init"

fix host_memory_profile doc typo

add ParallelDo example for benchmark/fluid

use apt install -y python-opencv to run flower dataset

* add normalize switch to box_coder_op

* add default prior box var

* update according to the review

add host_memory_profiling_cn.md

enable infer api with multi-threads

fix test error in fluid_benchmark.py

fix protobuf memory leak

Fix dangling pointer bug in paddle/fluid/framework/executor.cc

refine benchmark/fluid