* expose peak gpu memory API to python test=develop

* add unittest for peak gpu memory monitoring test=develop

* add pybind change test=develop

* add mutex to gpu mem usage monitor test=develop

* update benchmark flag definition file test=develop

* tweak unittest for memory monitoring test=develop

test=develop

test=develop

test=develop

test=develop

test=develop

2. Seperate the lock of kids and vars in Scope

test=develop

add unittest model containing while_op
remove unnecessary codes
test=develop

test=develop

* fix inference on gpu out of mem

the transfer logic in operator.cc will keep creating new scopes.

Fix code to support cpplint syntax check

test=develop

scope's API modifies its internal state. And scope's
API can be called from multiple threads during traing.
Hence, we need locks to protect the scope's internal
states.

We can optimize it in the future. But the current
solution is buggy.

test=develop

- add naive executor
- fix concurrency performance issue

* add memopt level

* add opt level for image classification demo

* clean code

* add delete op

* clean code

* test machine translation demo

* clean code

* clean code

* skip fill constant with force cpu

* clean code

* clean code

* refine code

* clean code

* fix bug

* "unify flags"

* "fix init"