Archive embedding size

[bjherger]

Archive vocab size, for use when creating embedding layer. Transient issue can occur where the maximum vocab index isn't seen in the training data set, and so the embedding vectorizer has a larger vocab than the embedding matrix.

Current state

• Embedding matrix pulls vocab size based on largest vocab index seen in training set

Future state

• Embedding matrix pulls vocab size from either transformation pipeline or somewhere else that is explicitly set by transformation pipeline