That was... fast! Thank you for taking the time to work on this.

The code looks a lot like Elasticsearch's code, so I suppose you took it from there and adapted it to our codebase? That's perfectly fine since their code is Apache 2 licensed, but we'll need to add notices to files that were copied from there, to preserve a "copyright trail". See for example org.hibernate.search.backend.lucene.lowlevel.docvalues.impl.NumericDoubleValues.

There are a few other things to iron out, like the code style and the tests that we need to add. I suspect we should also change the implementation of aggregations to use the sorted doc values instead of the old ones. Would you mind sending this patch as a pull request on Github so that we can discuss this and iterate? You can find a guide to sending pull requests in the CONTRIBUTING.md file; don't hesitate to drop by the developer chat if you need help.