Consider replacing usage of the brillig VM in `fold_constants_with_brillig` with the SSA interpreter

[TomAFrench]

We currently perform brillig gen halfway through the SSA pipeline so that we can continue processing some ACIR functions further while having the ability to execute any brillig functions for which the inputs are known at compile time.

We recently added the SSA interpreter which allows us to execute the SSA IR directly. This would allow us to avoid doing early brillig generation and just interpret the SSA in any situation where we're currently running the brillig bytecode.