The second one.
Just because a question is a duplicate doesn't mean it has to be phrased poorly. I'm saying this without looking at all into your asking history, so I don't mean to call you out as phrasing anything poorly, but that's often what a question ban indicates.
It's a well-known, and indeed oft-experienced, phenomenon when someone spends hours of their time researching a particular issue, but the solution never makes it in front of them. That's, ideally, what duplicates mean. That's why we generally don't delete them, and let them stir on as sign-posts to the original post.
If you're looking to improve a question which has been marked as a duplicate, the solution is the same as if you're looking to improve just about any other post: show your research, phrase it clearly, include examples, format it nicely, and you'll probably get some upvotes. They won't come nearly as quickly as they might on a question that isn't marked as a duplicate, but that's an inevitable tax that the implementation justifiably takes.
It's unlikely to happen, but it's possible that your post will be reopened and the predecessor will be marked as a duplicate of it. I say this, again, not to say it'll happen, but more to illustrate that a "duplicate" closure is not a death sentence. As long as you've done your research, and you show it in a well-communicated post, you'll be on the right track.
In response to your comment, looking just at this question, it sounds a little sarcastic. Particularly with the last line, it just sounds a bit accusatory. I don't personally think this question is deserving of this many downvotes, and in fact, there you go, have an upvote.
But for cases like this, it's often best to ask two questions, or just have two sections: first, establish what the best practices are. Once you do that, or if you can prove you don't have to, then it's a good time to question the system.
In other words, don't blame the system until you understand why the system works the way it does. At the scale of Stack Exchange, there are often considerations that come into play that aren't immediately obvious.
Beyond that, generally speaking, any time you tag something with feature-request, you want to make sure you have a well-defined grasp of what exactly you're looking for. Having discussions first to establish what the best route may be is often helpful to build consensus under less scrutiny, then you can move forward with something you know you have support for, and something that will generally be of higher quality.
As for your second question, about when we do delete duplicates, that's pretty rare in my experience. It only really happens if the duplicate is a very low quality post, or if it's an exact duplicate, generally verbatim.