Duplicate sub expressions (introduced during normalization) are problematic for "stateful" functions

[danmatichuk]

Certain functions get translated into "stateful" operations in macaw, e.g. write_mem or gpr_set. During normalization, functions are split into a distinct function representing each (sub)-field of its result type. This splitting is sound from a functional interpretation, but can introduce duplicate expressions if they are shared in the computation of each out put.

When these "stateful" function are duplicated, however, the resulting action occurs multiple times, and also is not necessary evaluated in the correct order.

This indicates that some structs should be retained in the normalized spec, which are exclusively used to manage the control flow of stateful operations. These don't require a logical representation in the expressions outputted by macaw, but instead would be translated into haskell-level tuples.