This has been a long debate throughout all of StackExchange, and been argued quite often on various metas. On main sites, votes indicate the usefulness/correctness of a post. But what to do on meta, where you have only one vote to spend and opinions regarding the quality and relevance of the post, and an opinion on whether it ought to be done or not?
The answers vary. Some people say "the vote arrows say 'This post is useful'! That tells is how to vote!" Others cry "read the tag! I disagree with the suggestion, however well-written it may be!" It is sometimes an arduous process to train users on the main site to vote down incorrect or low quality answers, and upvote quality questions. (There can even be debate over this!). And then once users are indoctrinated to the process on main, they move over to meta and make a quality, well-phrased feature request... That has been asked about a hundred times and The Powers That Be have famously declined. And in pour the downvotes! Upset users cry foul, asking what they said wrong in their question, and are flummoxed to be told "it's not personal; downvotes on meta signify agreement/disagreement. We just don't like the suggestion."
Hey... We never said it was a perfect system.
But in the end, what it comes down to is this: people have the right to use their votes however they want (so long as they are judging the content of the post, and not the user who posted it). We've given people the right to vote as they choose, and they're going to do so. The canonical answer is probably that yes, voting on meta is supposed to signify agreement/disagreement with a feature request. But you can't always count on all votes having been cast for that reason. People vote how they want to vote, whether they're aware of this longstanding debate or not. Give them a button, and they will click it. (Okay, that last was a bit facetious.)
In short: vote in the way that feels right to you, and take net scores with a grain of salt. There is one guaranteed point in favor of meta voting indicating agreement, though: when there's a popular feature request that the community is asking for and mods poke the Community Team, they request a meta post with 10 upvotes (not net score) to show that the community actually wants this. (At least on smaller sites like this.) So in that case we have actual working proof as to what's necessary :)
There are also upvotes on the answers to consider as you try and decide if the community is in support of something. Sometimes I'm indifferent on a feature request (wouldn't hurt, wouldn't help me much to see it implemented) and don't vote on the question at all, but an answer which tweaks the request slightly might earn my upvote (this interests me!) or my downvote (oh no, absolutely not like that). So... There's plenty of ambiguity to go around. I'm sorry if this isn't a satisfying concrete answer... I don't really have one. But I feel it pretty well portrays the situation as it stands, which is the best I can provide!