A binary representation `"\n" * 5 + "hh".encode('utf-8')` reproduces
the error from travis-ci.

* Add export for TensorFlow serving

Deploying models without the TensorPack dependency is sometime
useful or even necessary (TensorFlow serving). Example is also
provided.

* convince travis

* fix impotrs (tested)

* apply proposed changes to export

* use public api

* rename build -> export

* move location of export.py

* moving means cut and PASTE

* transpose_and_minmaxnormalize

transpose_and_minmaxnormalize

transpose_and_minmaxnormalize

transpose_and_minmaxnormalize

* fix linting