I don't think it's that simple. There is always a fine line between complexity and compatibility. The things that are barely used/almost unused are the things

I don't think it's that simple. There is always a fine line between complexity and compatibility. The things that are barely used/almost unused are the things that often end up the least tested or looked at. Depending on how complex they are, and what they do, this is a recipe for bugs/security issues. When you have thousands of these things, not removing them over time is a great way to end up with a huge mess. On the other side, as you say, removing any of them rarely benefits users directly (except in some individually minuscule way like binary size). What seems to happen here strikes me as a reasonable balance - people try to trade off the complexity of the feature vs its use, and wait for it to become unused enough that it might be worth removing. As you can see, here they decided it wasn't worth it, and it's still there today.