Feature request: support separate source and target vocabularies

[AmitMY]

At the moment, the vocab is trained on both the src and trg files. Instead, I'd like to have two separate vocabularies.

My use case is having two completely different langauges