Enhancement: support for pretrained word embeddings

[simonmandlik]

Implement a new Extractor subtype, called WordEmbeddingExtractor, for extracting NLP words using their embeddings (using Embeddings.jl and WordTokenizers.jl?)

Rough sketch of possible implementation can be found here, but this is for the old version of JsonGrinder.

A good starting point is NGramExtractor implementation, the design should be very similar.

We might also want to update suggestextractor with a new kwarg governing when Strings are extracted as ngrams and when they are tokenized