* Create Audio Feature in SDK
* Add apis to record audio in SDK
* Add corresponding apis in pybind, storage.py, sdk.h
* Implement reservoir sampling when collecting audio samples

* fix clang format and update based on comment

* use int_8 for reading records and convert string directly from vector

* refract isSampleTaken to IndexOfSampleTaken

* fix clang format again

* Add manual Save feature.

* Fix the incorrect style

* Add feature to record Text

* Binary Record Fixes

* Fixed issues with binary_record.h filename generator method.  std::hash is not consistent across OS.  It will generate different hash's when ran on different OS on the same string.  Change the filename generator to use a static counter.  It will be more ideal to use UUID generator available in boost 1.4.2, however, boost 1.4.2 has problems compiling on c++11 \n* Fix issue with pytorch demo where it logs to the absolute path /workspace

* Fix performance issue on server when server fails to load image file.  Currently server retries an API call 3 times before it fails, at which point it'll sleep the thread for 2 secs.  This becomes problematic when there are alot of SDK errors.  I'll create seperate issue to refine the retry logic to not be so expensive on the server.

* Add C++ and python style check to travis.  
* Update all C++/python code that violate coding standards.

Fix some typos

* init paddle

* add image writer

* change sample numble to 4

* add image and conv image

* add histogram

* rename test file

* link in eigen3 library to vl_test target
* Add new "template class SimpleWriteSyncGuard<Entry<long long>>;", since build will fail on OSX
* Add template keyboard to data<T>(0).Get() (otherwise this fails on OSX)
* Fix unit test in sdk_test.cc

Tested build and unit test run on OSX and Ubuntu with no issues

just a trick to store int8 in protobuf

should limit the maximum image shape latter