No, just NO.
Getting inf or -inf out of floating point actually makes sense: It is not just exactly zero but a whole range of very small numbers which give similar results and are treated as 'limits as x approaches 0'.
Integer numbers and arithmetic are exact and although INT_MAX might be large these days it is still not anywhere near infinity and just does not make any sense at all.
And having signed and unsigned behave differently also does not make any sense at all.
Your compile time option might as well be named '-foutput-random-instructions'