I've been building and using various forms of HTML transpilers for several years, as part of a build step or "JIT" during server-side rendering.
The main advantage I see is the ability to extend HTML to have layout/templating functions, building up a simple DSL for frontend devs and editors. Since the target userbase is already familiar with HTML syntax, it's easy to learn and gives them "super powers" - for example, an <include> tag that imports another HTML file/partial.
Another use case I've used in a number of sites/applications: chaining a Markdown parser, an HTML transpiler, and a React renderer for the HTML "AST". Among other things, it renders internal links <a> as <Link> components of the router.
And the rabbit hole goes deep: the XML/HTML syntax is actually able to represent "programs" in a Lisp-like manner. If the angle brackets are too noisy, one can use Jade/Pug syntax to generate the same AST.
In the end, the traspiler should render the result in standards-compliant HTML. This Pylon project doesn't transpile anything, so the result contains undefined tags which the browser treats as HTMLUnknownElement [0]. The bottom line though, is that it works - and I've seen numerous projects in the wild taking advantage of this behavior.
I'd also like to mention the posthtml project [1]. I haven't used it myself, but a generic HTML transpiler has great potential in my opinion.
Maybe we need to introduce HTML transpilers now :)