lower memory usage for facet sorted on unique(other_field)

Faceting on a high cardinality field and sorting by unique values in a different high cardinality field uses a lot of memory. We should implement an alternative method and pick the correct method based on index statistics and domain size.