Elementary questions are on topic, and in no way form a basis for closure. It doesn't matter how elementary a question might seem, as long as it fulfills our other criteria for being on topic.
Elementary questions are somewhat more likely to fail some of those other criteria - for example
they're more likely to be duplicates
they're more likely to have not been adequately researched (many basic questions are easily answered via Wikipedia, internet search engines and so on).
they're perhaps slightly more likely to be asked in an unclear or overly broad way
Such questions only fail those criteria by actually failing those criteria directly, not by implication. Which is to repeat, being elementary is not a criterion for closure, or even for discouraging a question.
Numerous discussions here on meta, in chat and in comments under elementary questions clearly establish that the general mood is to encourage well-asked elementary questions, not discourage them.
In short, gung is quite right in the original discussion. CrossValidated welcomes and has always welcomed supposedly elementary questions (while some people do have objections to elementary questions on various grounds, that has not been the general opinion any time I've seen it discussed).
Sometimes answering a good elementary question is a way that's both reasonably accurate and helps readers likely to wonder about an elementary but perhaps fundamental question can be extremely difficult.
This is one reason why I have frequently objected to labelling or characterizing questions as basic - whether by the OP or someone else in the body of the question, in the title, via tags or in some other way: the question might seem simple but that doesn't imply that a suitable answer is. Many deep and subtle questions are much easier to ask - so seem "simple" - than they are to answer.
So I won't define what's elementary or very elementary, largely because I don't think it's useful to do so. We don't need to draw the distinction except in so far as it informs the style of our answers (we want to reach our audience). That distinction can be entirely internal as one writes an answer ("I should answer this question in a basic way, and offer some intuitive motivation for it" -- if you want to see examples of that, I particularly recommend some of whuber's answers to basic questions).
Some users won't like dealing with very elementary questions, and that's absolutely fine. The easy option is to ignore them.
I think that even answering the elementary questions which have fairly elementary answers can be good practice -- finding better ways to explain the basic ideas is something that's an important part of our discipline, because statistical thinking doesn't come very naturally to humans (it's not something we have good intuition about). We need to do a lot better than many elementary texts do on this -- they frequently either fail to be correct* or they fail to provide any suitable intuition/motivation for a correct discussion (few beginner students gain much motivation from a purely algebraic argument for example).
I don't always choose to answer elementary questions, but I certainly don't see answering elementary questions as a waste of time (as long as they're not duplicates and so on). The overwhelming majority of the people who read the site -- including many who never post -- will not have much statistical background and so answers to the most elementary questions will have a much wider audience. My most popular answers are almost all to relatively simple questions (though usually with enough meat underlying them to require a fairly detailed answer). It's surprising how often I learn something in answering one.