Add option for local model, instead of downloading it every time.

[alfredo-f]

Hi David,

Pre-requisites of this pull request

This pull request is meant to be merged after #13.

What

This pull request add the possibility of loading the embedding model from a local folder, instead of having to download it every time you need to spin up the microservice.

How

• Add a new variable ARE_YOU_TESTING__LOAD_MODEL_LOCAL in utils.py
• Loads the model from a local folder with tensorflow.saved_model
• Add instructions in README.md

[daveshap]

I don't think this is necessary as TFHUB caches the model by default. This seems like superfluous code. It would be better to check in code at instantiation rather than set a static variable.

[alfredo-f]

I think this approach is a little bit more flexible:

• you can point to the folder where the model is downloaded once and for all, instead of having to rely on the caching mechanisms of TFHUB
• you can plug and play multiple models downloaded once, switching quickly between them to compare the performance