Import words from a training dataset, not (just) list of words

e.g. spacy only analyses POS in context of a sentence. For plenty of words e.g. `open`, the word can have different parts of speech in different contexts.

If we train on a bunch of pop music, we'll not only get more appropriate words to use, but also we'll be able to see the words in multiple contexts and import them again.

Will need to consider whether we update the schema to allow words to have multiple POS, or remove the uniqueness constraint of the word string and allow two versions of the same word, with different POS

Start by locating the current word-list import and the schema's word uniqueness constraint. Determine how a training dataset would provide contextual part-of-speech values and whether the schema should support multiple POS entries for one word; done means the import and storage design are agreed and implemented.

Written by the indexing model from the issue text.