When de Man constrasts a grammatical and rhetorical meaning of a rhetorical question, like "What's the difference?", what he means by grammatical is a "literal" meaning, one that asks for information. What he means by a rhetorical meaning is that the question makes an assertion: "it makes no difference, why are you bothering to even care about this."
So this is not really a distinction between grammar and rhetoric. Both questions have the same grammatical form and semantic meaning. Rather it is the difference between two pragmatic functions: one can use the question in that syntactical form to ask for information or to make a statement. Questions also serve as requests: "Could you open the window?" You could misconstrue the question as a request for information, and answer "Yes, I am able to open the window." That would be a pragmatic mistake. Questions can also be insults: "Are you really that stupid! What were you thinking!" Assertions can also be questions, as in typical interview formats. "You were born in Western Kentucky." In context that is asking the person being interviewed to elaborate on her origins. Or, "You're best known for your sestinas." The poet being interviewed might say: "Yes, I began writing them as a small child..."
It's not even a distinction between semantics and pragmatics, but between various pragmatic functions.
My point? Paul de Man's categories are medieval* ones. He wants to go beyond structuralist categories but he has little curiosity about actual modern linguistics, even. Grammatical is not a synonym for literal or non-rhetorical, in any case.
The alternative interpretation of Yeat's "how can we know the dancer from the dance" is totally dumb, anyway. It isn't plausible to have a request for information in this context, I'm sorry. It only works as a rhetorical question, in pragmatic terms.
*No insult here to medieval grammars, semiotics, poetics, and rhetorics, which are marvelously complex and interesting.