Refactor Label Inference and ITE Assignment

[dz333]

Currently, in large programs with lots of inferred labels, the ITE labels can get extremely large before being simplified. For example the value of a label may be inferred to be:

( var1 ? BOT :
              (var2 ? BOT :
                                     (var3 ? BOT : pc_lbl)))

However, it is often the case that var1 == var 2 == var3

In the currently implementation, this is resolved after creating these deep ITE labels and causes large recursive resolution that is inefficient. This has been solved in JOIN label inference by sorting sublabels; however the same approach cannot be used with ITE labels since their sublabels cannot be extracted since ITE is not a commutative operation (the join optimization just pulls out all labels referenced in the join and rejoins them later in sorted order).

Instead, ITE assignment must be done after some inference has resolved these equivalent labels to minimize the depth of ITE labels necessary to express accurate flow restrictions