Sure, contributing to link rot is bad, but in the same way that throwing out spoiled food is bad. Sometimes you've just gotta break a bunch of links.
That probably depends on the link's purpose.
"The full dataset and source code to reproduce this research can be downloaded at <url>" might be deeply interesting to someone in a few years.
In any case a paper should not rely on an ephemeral resource like internet links.
Have you ever tried to navigate to the errata corrige of computer science books? It's one single book, with one single link, and it's dead anyway.
There are always dependencies in citations. Unless a paper comes with its citations embedded, splitting hairs between why one untrustworthy provider is more untrustworthy than another is silly.
This is by no means a universal experience.
People still get printed journals. Libraries still stock them. Some folks print out reference materials from a PDF to take to class or a meeting or whatnot.