Quick-and-dirty answer: antithesis

The skinny:

Keep in mind that a taxonomy, e.g a list of rhetorical devices, is imperfect and descriptive in that it artificially breaks down complexities into more basic concepts purely for didactic or organizational effect. A common misconception is that biological taxonomies are telling of evolutionary development, like a family tree. They aren't. In them are grouped together organisms that have similar characteristics- in preponderance. Occasionally we will hear about a re-classification of an organism, not because of any new historical information, but just based on a reconsideration of the observable features.

In that sense a classification of rhetorical techniques is akin to a model, which is a simplification of a real-life phenomenon. A model allows a point of departure for analysis, uniformity in descriptive terms, but remember the old saw, "don't take the map for the terrain".

Having said that, the phenomenon posed may also be understood through several descriptors at the same time, showing characterisitcs more of one than another, etc. Should we classify it a pun or as another contributor suggests paraprosdokian. I would add that it has traits of antithesis as well. Which is the overriding characteristic that will determine the concept's classification, but at chance of oversimplifying it?

This is basically a long answer to say that as far as I have researched it, there is no precise label. The known classifications are not as specific. A sub-set label would need to be coined (at the risk of inordinate rarefication as mentioned).

I would say the phrase's overriding distinction is antithesis. At the same time, that the two concepts are distinct is not immediately obvious, indeed creates a surprise, sometimes humorous effect. That the distinction is so slight, the listener perceives the keen observation of the author, his or her sophistication, then by which we could tag it 'wit' or a 'witicism' (which is not a classical rhetorical device of course) I think it is a pun as well, a pun being a very wide-encompassing category.

At any rate rhetoric happens, it is a feature of human language that is not restricted to speech-making and essays as some posters here might suggest. It is the 'how' of a message, particularly the 'how' a message influence some desired outcome. It doesn't have to be deeply premeditated to be what it is.

Many text-book examples for the devices are indeed 'paradigm cases', but just as many on closer inspection reveal characteristics of other devices as well. The question becomes one of preponderance. In that simple classification something is lost, however, and in the examples given, too much would be lost to suffice a sole label. That is, of course, because none of us are aware of a specific named sub-set for the examples you have chosen. They are you might say the paradigm cases for a yet-unlabeled phenomenon. I think that means get to name it ;)