Accept callback_free(0) instead of bugchecking

callback_free scans the thunk table for the supplied pointer and __fastfails
when nothing matches, so callback_free(VAProc:0) -- the natural thing to
write after a failed callback_alloc -- brought the machine down. pawn/test.p
escapes it only because main() bails out when callback_alloc fails and a
failed load never invokes unload.

Treat 0 as a no-op. Every other bad pointer, including a double free or a
slot owned by another VM, remains a module bug and keeps its __fastfail:
there is no way to report it that a correctly written module would ever see.