OpenVINO Model Serving app

[benhoff]

Reference: #759

In the future we need to have another solution. It seems that internally CVAT doesn't know about possible inputs of a specific model. Need to get the information from its interpretation script. In the future the interpretation script can have a specification of the model: prepare input, process, handle output, etc.. Look at spec_info for annotation formats which we introduced recently (https://github.com/opencv/cvat/blob/develop/cvat/apps/annotation/README.md).

Next step is to integrate OpenVINO Model Serving (https://github.com/IntelAI/OpenVINO-model-server) and Tensor Flow Model Serving (https://www.tensorflow.org/tfx/guide/serving). It is more general approach to work with DL models.

1. create a tensorflow model serving app

• Server: https://github.com/tensorflow/serving

2. refactor the existing or create a new OpenVINO Model Serving app
3. create a smarter preprocessing step as part of the interpert script