> JS improves the user experience a lot, both in terms of interaction and speed, as well as making development more manageable
Almost. Most JS (caveat Google Docs etc.) creates the kind of experience users have come to expect. That's orthogonal to "improving" the user experience. Because of the keeping-up-with-the-joneses effects of frontend engineering and design patterns as spearheaded by giant, framework-authoring, standards-setting companies, there are strong incentives to converge on bad outcomes. For example, consider the common trend of replacing built-in controls with equivalents made out of "div soup" for the sake of providing consistent styling. This is done because it is thought that styling controls is a table-stakes necessity for interactive websites. Many studies indicate this, and many users request/demand it.
It's also misguided.
That's not to say that the studies--or the users--are wrong. But "this is how it is done and this is how we want it" is a norm, not some objectively superior optimum, and it's a norm that was created by many years of development. Some of that development was necessary, and advanced the state of the art. But much of it was fad/hype-driven, poorly thought-through, and slipshod. The end result is a bunch of tools/patterns/user expectations that bear little relation to what actually improves user experience.
In other words, JS often only improves things if you grant that the currently accepted definition of "improved" is actually an improvement over using other tools; often, it is not.
> at the expense of annoying purists
In other words, engineers. Engineers more interested in delivering quality than complying with norms or maximizing marginal click revenue by adding the 157th tracking library to a 50MB bundle.
> CSS/HTML gymnastics
Engineering.