Speed up [] grouped by a single annotation

[jan-niestadt]

If a user tries to group the any token query [] by a single annotation, right now this is resolved using HitGroupsTokenFrequencies. This is faster than finding hits first, then grouping, but it could be done even faster by the method used by the /termfreq operation, which simply sums term frequencies from Lucene's reverse index. It would be nice if this path were automatically chosen in this case.