Here's the guidelines as I see them.
Avoid changing a question's meaning
To begin with, you're absolutely correct - once a question has been written, it can be edited (Stack Exchange gives users a lot of authority, including proposing edits to other people's posts), but its meaning should not be changed. For precisely the reasons you state: people have invested time and effort into posts, and you can't change the meaning of the question without invalidating existing answers. A question undergoing substantial revision finds its answers ceasing to make sense.
So this is true whether you are a moderator, a random community member, or the person posting the question to begin with. Once you've asked the question, that's what the question is. You can clarify, but you shouldn't change -- if you need something new answered, the correct thing to do is open up an additional question.
The problem is questions that are ambiguous to begin with.
What is an ambiguous question?
Some questions are ambiguous -- they could be interpreted in several different ways, or nobody's sure how to interpret them at all.
Some typical examples:
- Questions that ask more than one thing. Otherwise you get people answering one question and not the other, and then trying to vote and compare between two answers to two different questions.
- Questions that are overly broad or unclear. If a question isn't really clear or specific enough to be answered, than by definition, attempts to answer it are likely to confuse issues even further. Answers can't match the question if you don't know what the question is.
- Questions that participants are misunderstanding and answering the wrong thing. Maybe the OP thought he was being clear, but people keep latching on to some side detail and understanding that as the real question. Again, we have answers that are trying to match the question, but missing the point.
- Questions that get answers that don't seem to match the question. Even if the question looks kosher, if the answers you actually get seem not to be addressing the same question, then you've got a problem. Whatever the reason, if the answers don't match the question, there's a problem here to be resolved -- maybe it's one particular poster; maybe it's a larger misunderstanding.
And here's the thing: All those problems we mentioned with questions being changed after they're written? With ambiguous questions, they're baked in at the start. Because if questions and answers are mismatched, then you've got people spending effort on answers that aren't helpful, and people posting and getting answers that don't address their question, and good content getting deleted because eventually the question does get focused (or closed).
So in these cases, "Don't change the question's meaning" is irrelevant. The question doesn't have one single meaning. The question needs to be fixed -- and that might mean invalidating some answers.
Don't answer ambiguous questions
This follows from what we've seen up until now. An ambiguous question is not aided by a good answer. An ambiguous question is aided by closing it, putting in on hold, commenting for clarification, and doing whatever it takes to remove the ambiguity.
If your answer has been invalidated, re-post with an appropriate question
There are still hiccups, and sometimes good content gets written that's mismatched with its question (or where the question gets closed).
The simple solution for these cases is: ask a good question, that is matched to your answer. And post your answer. Stack Exchange encourages self-answered Q&A. This is an awesome way to keep your work, and make the site better.
In conclusion, let me say a few words about the particular case at hand.
In this case, we had an ambiguous question. It could be read as per the title, "What can you say when there's not a lot to say?" -- a meta-question about writing literary criticism, using two particular poems as examples. Or it could be a very specific question -- not "How would one effectively comment on writing in this specific context (of not having anything to say about it)?", but rather "What comments can be made about this specific poem?". To be blunt, I read the first draft as clearly being the first of the two -- not asking specifically about "To Autumn" at all; merely using it as an example of a type, of poems that have beautiful imagery but little depth.
As I said, your analysis is great, and good reading. That does not guarantee that it is an answer to the question.
Please consider: If you had, hypothetically, misinterpreted OP's question (as could be assumed, if his clarification really had rendered your answer was irrelevant), what would that actually mean? It would mean telling OP he can't have his question answered, can't get an answer to what he actually asked, because you misinterpreted it and spent effort on a misguided answer! That's not any good either.
Instead, we rely on community moderation (that is, everybody experienced and active enough participates in moderating the site), on discussing issues as they come up, and hopefully on a healthy dose of good will.
In this case, the original question was poorly framed, but you interpreted it basically correctly, and the question can be edited in a way that preserves your answer perfectly. In another case, a different solution could have been found -- such as splitting off to a new question of your own. And in all cases, we're trying to discuss issues respectfully, looking for whatever solution will leave the question, the content, the site, in the best condition possible. In the long run, that takes precedence over any one post.
I hope this answers your question. From my experience, ambiguously-worded questions are a significant issue on "soft" Stack Exchange sites. So it goes :)