Ok so, what about "duplicate" the data over two indices like "whatever-index" and "whatever-index-unique"? On the former you could put all your data (useful for drilldown purposes) while on the latter you basically use the product_serial as document_id and simply don't ingest a document if a doc with that same id is already in the index.
That way you could make an index pattern like whatever-index* and in the visualization that requires document to be unique you filter in the search bar by _index:whatever-index-unique.
Might that be an idea to simplify whatever you have to do?