* Batch norm folding

* Addressed PR feedback

* Style fixes

* Style fix

* use reference kernel for reverse_sequence for int

* move tests

* resolve CI errors

* TEST to NGRAPH_TEST

If user manully provide MKLDNN_INCLUDE_DIR and MKLDNN_LIB_DIR, don't
build mkl-dnn and just add a dummy external project "ext-_mkldnn" to
satisfy target dependency for the rest of the build.

* give frontends some flexibility over fusions they would like to run

* address jbobbas feedback

* Make sure that generating pyapi does not throw errors due to directory structureor linenos

* Update basic.py

* Output warnings and compiler errors from NVRTC, updated avg_pool kernel for cuda 9.

* Added defgaurd to only apply defines for deprecated function if CUDA_VERSION < 9000. Updated shfl_xor invocation.

* Make sure that generating pyapi does not throw errors due to directory structureor linenos

* update docs

* one blank line instead of two

* blank line error resolve

Move temporary buffer pools to CPU runtime context to avoid static variable destruction issues with clang (#990)

* Enable NNP reverse build, clean up ngraph repo

* clean mkldnn cmake

* add reduce op

* fix bug

* fix bug and enable tests

* hack solution to get reduction function in reduct op

* hack version working on all tests

* fixed the reduction checking process

* clean up, rename

*  remove commented code; add comments

* throw useful error when Backend library fails to initialize

* add dlerror to error message when opening backend shared lib

* Generate binary representable floats for input data

Removed ReLu activation present just before softmax.
(It's not a good idea to have ReLu > Softmax immediately. The test accuracy of the model is now much better.)

* move nop elimination pass to nGraph and add broadcast elimination

* fix pad test bug

* remove graph visualizer and clean up includes in nop eliminate test

* code format

* Final PR review edits plus repair abc.cpp example docs that broke when code was added

* Word

* type tests for reverse_sequence

* remove commented out code

* test_control in util

* Add mechanism for disabling specific backend unit tests from a manifest file.
Populate the test manifest files for CPU, GPU and INTERPRETER.

* update docs for new manifest controlled transformer unit tests

* [Py] Use get_friendly_name for object names

* Moved emit_elementwise implementation into CUDAEmitter and added logical_and and logical_or ops.

* Updated comment and formatting.

* Added check for multi-output elementwise ops.

* Added op::AvgPool cudnn impl. which works for 2-3 spatial dimesions and no/symmetric padding. Enabled tests.

* Added cuda-c implementation of average pool which handles 1-3 spatial
dimensions as well as asymmetric padding. This commit also introduces
several helper functions for performing fast integer division and
fast constant memory access.

* Formatting. Removed bool that was used for testing to force the cuda impl. over cudnn.

* Added CUDNN AvgPoolBackprop implementation.

* Removed inline enum in preference of a helper struct. Removed instances of multiple declarations on a single line. Updated comments.

* Removed _prefix to helper functions in anonymous namespace.

* create ngraph_test_util

* installs libngraph_test_util