Improve performance of importing labels

[saumier]

There are 2 SPARQLs that load missing labels for objects, qualifiers and references. Currently the labels are all dumped into the default graph. The SPARQLs are very slow. The "load labels" launches seperate SPARQLs for objects and for qualifiers/references. They took 20 minutes for 3 productions + tertiary nodes.

I propose saving labels in a named graph that is shared, and optimizing the SPARQL by removing some filters and adding more performant filters to minimize the calls to Wikidata.

For example improving the removal of entities with existing labels in the default graph (or other named graph) before doing a federated SPARQL to Wikidata:

  filter not exists {
          graph <http://www.ontotext.com/explicit> {
              ?o rdfs:label ?b .
          }
      }