coverity: incorrect model of `stgMallocBytes` and `stgReallocBytes`

The Coverity 'model' of stgMallocBytes (in utils/coverity/model.c) is incorrect: it claims the function never returns NULL (assert(mem != NULL)), which (in its current implementation) is not true (cfr. #22380). Hence, the static analysis could make incorrect assumptions and miss certain bugs.

Similarly, the model of stgReallocBytes seems incorrect, as in, it claims to free p and return NULL when n equals 0, which is correct for Glibc's realloc (though non-standard), whilst the actual implementation would bail out in such scenario (realloc can indeed return NULL, but stgReallocBytes doesn't like that, in any scenario).