Store pre-aggregated data in sparse indexes

Our sparse indexes record the min/max value on multiple levels to speed up filtering. We could store pre-aggregated data there as well to speed up faceting, e.g. sum and value count. The main problem to solve there is how to make `Collector`s take advantage of this information at search time.

Start by tracing how sparse indexes record min/max values at multiple levels and how Collectors use index information during search. Determine how pre-aggregated sum and value-count data could support faceting, then define what Collector behavior and search results would demonstrate that the work is complete.

Written by the indexing model from the issue text.