I am not a native speaker of English, but I studied in the UK and lived there for a long time.
I'm a native speaker of US English, and these nuances of unusual agreement are areas known to differ in US and UK English, so that's worth being aware of in the discussion. I can't comment on UK English, and it might be different from what I'm saying (or even narrower dialectal variation).
To me, at least in the British usage "Vodka and water" refers to a cocktail. Otherwise one would say "Either vodka or water..." in your example.
True, confusing example. It's hard to come up with a good one for what I wanted, but just assume the context makes sense. Like I said, where you're trying to avoid drinking too much alcohol without water. Or if you go to a party and they're offering both (and only) beer and wine, and most people have a preference and drink only that, but then you say "Come on everyone, beer and wine is best!"
However, I think it's important to clarify where one comes from in terms of orientation. I am absolutely anti-chomskian.
I've been trained in a Generative framework, but I also find Constructional approaches to be useful. In my research I'm looking into ways of taking the most important aspects of both. (I don't think it's an either-or answer.) Admittedly it's much easier to find fault with both theories than to be sure about which aspects of either are definitely correct.
To me, analysis starts from what people acutually say and not from what they might say but never do.
That's complicated-- in principle I agree with you (and reject 'armchair' approaches to things in favor of gathering real data), but it is also important not to overlook more peripheral types of usage in favor of only common expressions. And this is actually where Construction Grammar excels!
Personally the biggest problem I have is that I think many armchair style Generative analyses are not clear enough (or imaginative enough) about the contexts in which an utterance might be used, and therefore miss generalizations about grammaticality. I'd rather think of it not as whether a sentence is, outside of context, grammatical, but what it would mean if it were to be used. So something like we're describing here does have a clear intuitive meaning to me, even though it's a very rare sort of expression. (But as a whole, certainly not unused. Just rare enough it's hard to think of specific examples. Searching in a large enough corpus would surely generate some results under the right parameters! This isn't just hypothetical.)
I agree with "wine and steaks is good", but then again, to me it implies an omission along the lines of "(a meal with)...".
I don't understand your insistence that there must always be a source construction (rather than as one potential analysis). As I've said, I believe it is citing a grouping as a general concept, which is in that sense not countable, and therefore gets singular agreement. It's grammatically like a direct quotation, shielded from external grammatical relations.
In this case, "a meal with" is not the appropriate substitution because it is a more basic concept than that. Maybe it's a snack. Maybe it's a smell. Maybe it's a look in a photo. Maybe it's a recipe. Maybe it's a taste of a meal but not a full meal. Or if you change those two items for some other combination it might be any other number of things. Just because you can paraphrase it doesn't make it ellipsis. It may be analogical to diachronic ellipsis (as you made a case for earlier), but not every combination like this needs a specific source construction.
Otherwise how would justify it?
That's easy! It's just a concept. Like an infinitive form of a verb with no tense, this is an "infinitive" form of a noun phrase, with no number. It's not
being counted, so the grammar isn't sensitive to that either. It's a concept, like "mac and cheese" or "steak and eggs". It's not a set phrase (binomial) like those, but it works the same way because "wine and steak" is a
pairing rather than a plural group of two. If you really need it to come from ellipsis, then I suppose you could claim all of this is something like "[a set of] wine and steak", and the
set is what gets singular agreement. But I just don't have that intuition, although I admit that such an argument isn't really testable or falsifiable. Either "ellipsis" or "abstract concept" would give us the same data.
The reason for my strong intuition is because of minimal pairs like the following:
"Red wine and steak is delicious."
"Red wine and steak are delicious."
Those have very distinct meanings to me, exactly as would be predicted from what I'm saying about the two distinct analyses. The second is sensitive to grammatical number specifically because it is semantically different from the first. The second is a list. The first is an idea.