We currently call estimateRIDs twice per ixscan node: once to estimate outCE, and another time to estimate inCE. 

This can potentially be made more efficient by somehow integrating the two estimates: eg. iterate the sample only once: for each document in the sample increment inCE if it matches IndexBounds, and also increasing outCE if it also satisfies the residual filter.

This would let us halve the number of indexbounds comparisons.