Currently, an instance of Lucene's Document class is used to represent index changes and is passed to index managers etc. This requires the Elasticsearch backend to do quite some tricks to asemble a JSON structure representing the document contents, specifically when it comes to nested contents.

Hence we should find a more abstract representation of indexed "documents", allowing different backends (Embedded Lucene, Elasticsearch) custom implementations. Amongst others, the FieldBridge contract will need to be changed (or a new, alternative form be added) to work with this more abstract representation.