Mutable dependency tracking for normal variables

[SSoelvsten]

Zhang and Kifer [ZK17] assume all free variables in a normal variable are immutable. Currently, we do no such check.

One wonders though why they do this assumption. Based on their description of how to handle random variables and their proof of soundness, it seems it all is connected to whether the distance of a variable x merely is the inferred value or one has to use a ghost variable ^x to contain the backup of the old value before mutation.

One should hence think to extend the distance tracking of #8 to also include normal variables.