This is also makes "zombie CSS" a thing of the past, since the resultant CSS file that tailwind generates is based solely on what is in use.
Write a class & map it to the relevant component or DOM node in a template, so you only have a single source of truth to maintain the styles of that component.
Writing the explicit style declarations directly to the DOM nodes themselves is precisely what _prevents_ portability and reusability.
The perf issue of "zombie css" is seldomly an issue in my experience: if it really impacts your load time or perf, you can still easily serve subsetted stylesheets at build or request time
If the perf impact is negligible, you might in turn profit from serving all the css at first, since it will be cached and accelerate subsequent request or rehydration
Removing unused CSS is mainly a concern for _inline_ styles, since the bloat of the initial HTML might impact FCP and repaints/reprocessing once additional stylesheet are loaded.
... which ironically also means that adding truckloads of utility classes to every DOM node might introduce quite some initial load & paint performance implications aswell, depending on your app.
It might even be interesting to benchmark if a stylesheet with a bunch of "zombie css" really has more perf impact than serving a minmaxed css file, but requiring dozens of classes on every other DOM node to be processed for painting
it's not a file size game in the end, the amount of selector statements & DOM nodes they might apply to has much larger paint perf impact than a few kb of unused gzipped text