When computing the productivity of plans after the estimation trial run during cost-based ranker selection, use the document/works counters of the child objects under LIMIT and blocking operations.

This will address the usual case when the parent will not produce any result documents despite its children having pretty good productivity simply because the estimation trial is too short. Currently the productivity is 0, which is wrong.

The analysis of plans should take into account that LIMIT may not be the root node in general (or at least check if this is the case for CBR).