Store embeddings in different file format

The embeddings are stored as pickled files. Pickle has security issues and can result in backwards compatibility issues. 
The embeddings are now stored as pandas dataframe. 

A few options are feather or parquet. These seem to be fast, but I am not yet fully sure about the backwards compatibility. 