We currently always add a filter to Lucene queries, to make sure only "root" documents are returned.
This filter is obviously unnecessary when there are only "root" documents because there are no "nested" @IndexedEmbedded.
It turns out we can get a small, but noticeable performance improvement for MatchAllDocs queries by removing this filter. It also brings us closer to Search 5 performance for this particular type of query.
Tested with an Infinispan performance test: