Add weight normalization wrapper.

add docstring for set_gradient_clip

* initialize batch barrier

* add some comments

* update

* fix batch barrier

* use sendvariable rpc interface to send batch barrier

* fix comment

* fix method

* fix by comment

* fix by comment

Add build_type to build.sh

Feature/Recv op python with guard

* Remove IsBounded as buffered channels have to be bounded

* Add derived classes Buffered and UnBuffered"

* Implement buffered and unbuffered channels

* Correct the syntax of Channel::Receive

* clang-format

* clang-format 3.8

* clang 3.8

Add python api for lstmp operator

* fix typo

* fix typo

Add DataTypeToString

Update copyright for notest_dist_image_classification 

Fix the decoder_boot in test_rnn_encoder_decoder.py

* Polish threadpool

* Add #include <vector>

* Rename variables

* Rename variables

* clang-format

* fix some typos

* fix

Downgrade boost to fit manylinux

Fix send op data race

Make notest_dist_image_classification  consistent with distributed implementation in others. (#7899)

* Make this file consistent with others

* fixed style

* add Channle

* refine Channel

* Add one_hot operator.

* Add more unit tests.

* initial commit

* add new executor run function

* fix bug

* fix multiple definition of feed_fetch_method issue

* fix cmake

* fix tensor copy error

* refine executor code

* add comments

* temporary modification

* address comments

* fix bug

std::vector<bool> is not safe for concurrent write, even to difference indices.
More discussion:
https://stackoverflow.com/questions/48452611/is-stdfuturewait-a-memory-barrier-i-can-not-explain-this-data-race