Inconsistent executions in allow-set due to deadness predicate about memalloy HOT 2 CLOSED

We should try to remove this particular constraint on deadness. This will require changes to the generation when we have a control-dependency from a read that reads the initial value 0.

Two possible ways of dealing with this:

• Change the initial value of the destination register (say, -1)

• Change the assignment of the destination register into a ternary:

    rd := rn == 0 ? load(x) : -1
  

from memalloy.

I've kept this deadness constraint, but now it only appears in the deadness definition for C++/OpenCL models. Hardware models no longer enforce it.

from memalloy.