The "everything already in it" part is the bit doing the damage, and it is worth being blunt: those months are gone whatever you decide. They are not an asset in the project, and continuing does not recover them. The only question available to you is what the next three weeks are worth, compared with the next three weeks spent on anything else.
A test that helped me: if a friend described this project to you — four restarts, no features, months of maintenance, still excited about the idea — what would you tell them? People are consistently able to answer that about someone else and unable to answer it about themselves, which tells you the difficulty is not analytical.
A second one, which is harder: would you start it today, knowing what it takes to maintain? If the answer is no, you are not choosing to build it, you are declining to stop.
One thing I would push back on in your framing, though. You say the idea is still good. If that is true, the idea can survive the codebase dying. Ideas are not stored in repositories. If you genuinely want to build this thing one day, killing four years of accumulated rot and starting clean when you actually have the time is a completely reasonable plan — and it is a different decision from another restart, because it does not commit you to three weeks of repair first.