[m-dev.] invalid preds

[Zoltan Somogyi]

Would anyone object if I changed the compiler to stop and exit
after semantic analysis if the set of predicates marked invalid
by some pass by then is not empty? It would allow us to both
simplify and optimize later passes that at the moment need to pay
attention between the distinction between valid and invalid
predicates, even though the work they do will be thrown away
when the compilation as a whole fails due to the program bugs
that caused predicates to be marked as invalid.

Zoltan.