Keep it & enforce it better
The tag gm-techniques is one of those that arose organically from the community. I think it does have value and should be kept as-is. But I agree that it gets used too broadly, and that dilutes its value. If we can possibly identify other existing or missing tags that would cover the current mis-uses of the tag, we should more aggressively retag with them.
The gm-techniques tag is currently the dumping ground for "um, this is about GMing, I guess this needs a 'GM' tag of some kind" questions, and that's not useful.
What should [gm-techniques] cover?
I think gm-techniques should only be applied to questions that are specifically about concrete applications of the skill of GMing, which is a literal interpretation that matches most definitions of "technique", especially these two:
1. the manner and ability with which an artist, writer, dancer, athlete, or the like employs the technical skills of a particular art or field of endeavor.
4. technical skill; ability to apply procedures or methods so as to effect a desired result.
Only if the question is very pointedly about or asking for procedures and methods to achieve a specific result using the technical skills of GMing, should it have the gm-techniques tag. (Extra bold on "technical" is deliberate.)
If a question is asking for general GMing advice, then the tag is not warranted. The answers might provide techniques or might not—but the question is not specifically about techniques, any more than a game-rec question is about whichever systems are recommended by the answers.
As a rule of thumb, to qualify as a technique, something has to be asking both "how" and be about the application of the GMing art and theory to a certain end. People will often assume that any question that starts with "how" is asking for a technique, because one loose synonym is "method", but not all "how" questions are actually asking for techniques.
To apply this to the categories you've identified (IMO of course):
questions about improvement
Learning specific techniques is only one way of improving one's GMing. (Others include educating oneself about the division of authority; correcting misapprehensions about what GMs do; learning how to understand and apply the rules of one's chosen game; etc.) Answers might provide techniques, but the question is not asking for specifically or only methods of applying one's GMing skills—if it is, it would fall under the "literally about techniques" category.
questions about theory
Theory is by definition outside the scope of techniques—techniques are, by definition, applying already-established theory. Theory and techniques go hand-in-hand, but are very distinct. If a question is about both theory and technique equally (not one being background to the problem) it is probably a "double question" that should be split.
questions about communication
Sometimes this will be asking for specific communication techniques, like how to pass secret messages to players. Then yes, of course. Sometimes this will be describing a miscommunication and asking what the heck they should do. Again, answers may provide techniques, or might not, which does not make that question itself about techniques.
questions about dealing with players
Most often this will not merit the tag, as these are usually about social issues, and the problem and its good solutions will cover much broader territory than mere tricks of the trade. Only the rare question will be specifically about techniques; questions about how to speed up players who are slow to take their turn, for example, would almost always be asking for GM techniques (specifically the time-management subset of GM techniques).
questions about tough GM situations
Sometimes this will be a focused question about a specific problem due to lacking or failing at a specific GM technique. Much more often though it will be about wild and wooly group issues like playstyle conflicts, finding or removing players, and the like; which aren't about GMing techniques so much as about "being a social human" techniques.
questions about pink nerve-gas farting dragons
Despite the outlandish name and catch-all nature of this category, these are often about specific techniques. Whether and when to fudge dice, how to convey GM-secret information for the desired effect, how to shape a game to encourage certain types of PC behaviour—those are all technique questions. (How to craft good villains will be, depending on the specific problem, either a GM technique question or belong to gm-preparation.) Sometimes they'll be more social or personal in nature though, and be broader than specific techniques—avoiding burnout and disillusionment are that kind of problem that aren't likely to be laser-focused on techniques. These are "how" questions, yes, but they aren't about applications of the GMing art, but broader being-a-human issues.
What tag(s) should be used for everything else?
A few good existing tags for GMing problems that don't necessarily involve techniques:
Possible new tags we could use:
communication (13 letters) for "soft" questions about having communication problems
communication-techniques (24 letters) for "hard" methods-seeking questions that are more focused than "soft" communication-problem "what now?" questions.
(Although as always, too few tags is better than creating unneeded tags.)
For example, the question "Running games for a constantly unreliable group" could probably be tagged gm-preparation group-dynamics encounter-design. It's not asking for specific techniques, but more for how to structure sessions in order to reconcile the conflicting needs of the group's dynamics and the game's internal balance mechanics. That's not an application of the GM's art during play, but rather how to execute one of the GM's many out-of-game responsibilities: the broad problem of logistics and how to adjust the game rules to accommodate.
Similarly, our classic question, "How do I stop my PCs from acting like insomniac monkeys on crack?", is more about social-contract, problem-players, and maybe group-dynamics. Notice how the accepted answer only briefly touches on a technique ("Go around the table..."), while most of it is about communicating, adjusting expectations and social contract. One of the pieces of advice provides a chunk of theory ("give them useful choices") and then an application thereof—a technique—a tidy example of how the solution is a mix of technique and non-technique, meaning the question is itself not specifically and only about GM techniques.
Another example, "Is it fair to press the players for decisions during combat?", is an interesting case. It is explicitly about a technique they're already using, but the real problem is about whether it's fair to use that technique or not. This one could probably take or leave gm-techniques, but almost certainly should have social-contract on it as the most important tag, as that's the best-fit tag to categorise questions about fairness.
But then we won't be able to search for all GM-relevant questions!
But wait, that means that some questions won't have the letters GM in the tags anywhere! That's... an interesting result. It suggests that we might want to have some way of collecting all questions useful to GMs together, even when it's not about specific techniques. One way of doing that would be to resurrect the gm tag as a marker that "this is relevant to GMs"... but that has known problems.
So maybe we don't need a "this is for GMs" tag at all. After all, GMs are disproportionately represented online, so a huge fraction of our questions are relevant to GMs. In the same way we don't need a [rules] tag because it would apply to a ridiculous number of our questions, possibly we just need to trust that GMs can handle having to look through more specific tags if they're browsing tags for advice. Got group issues? Read groups. Got prep issues? Then look at gm-preparation. Got both? Well, instead of wanting to be to be fed a huge self-defeating firehose of all GMing content on the whole site, put on your big-kid pants and deal with having to look in all of two places to find targeted content. ;)
I think the latter approach is growing on me. We don't need an "all GM questions" tag. If players can find content just fine without having every player-relevant question tagged "[player]", then so can GMs.