In conditional probability queries - refine directly preimage of conditioning Rand Var

[zenna]

Currently we do two preimage computations - one for Y and one for X & Y. Both of these preimage queries refine starting with the whole of Omega. A better way might be to find the preimage of Y first, then use that preimage as the starting point for the refinement of X. We get an intersection of X and Y effectively by construction we may help with

1. Speed we're doing less work so it should go faster
2. The problem of correlation when dividing two intervals for computing the conditional probability

Being able to start refining from an arbitrary subset of Omega will be good for a number of other reasons. In particular we can refine a bit, if the result is satisfactory we can refine some more.