Make query construction more reserved to reduce unnecessary overhead

There is no need to filter for the root of a facet as every dataset should be attached to all facets, i.e. at least via a "other" dummy category. Hence, we only need to add the actually relevant sub-queries to our top-level. Furthermore, if there is actually only one sub-query, i.e. the one determined when the user types in their query, we do not even need the top-level intersection thereby simplifying Tantivy's internal weight construction.