When examples are given, they typically are with scalar values where such behavior makes sense: the resulting scalar value has to be NULL or non-NULL, it can't be both.
It is less sensible with compound values where the rule can apply to individual scalar components. And indeed that is what Postgresql does for another compound type:
I agree completely. Scalar vs compound structure seems like the essential difference.
You don't expect an operation on an element of a compound structure to be able to effect the entire structure.