* Explore: facetting for label completion

- unified metric and non-metric label completion
- label keys and values are now fetched fresh for each valid selector
- complete selector means only values are suggested that are supported
  by the selector
- properly implemented metric lookup for selectors (until the first
  metric was used which breaks when multiple metrics are present)
- typeahead tests now need a valid selection to demark the cursor

* Fix facetting queries for empty selector