Check external links in CI

[straight-shoota]

We previously had a CI workflow to check external links. But it was putting a big load on CI response time and was mostly ignored anyway.
So we dropped that in #921.

Link checking is notoriously difficult. It's a huge task with thousands of links, and it's quite common to see some temporary availability issues here and there.
Also, an increasing number of hosts rejects automated requests to protect against overburdoning scrapers.

This issue is to find a better solution.

We could consider moving this to a separate workflow, maybe on a schedule and not allowed to fail. Then we'd at least know when something is broken.

It could also be useful to check links already in Markdown (https://lychee.cli.rs/ can do this) and only in files that are changed in a pull request. This ensures that links are at least valid at the time of creation.

[straight-shoota]

#957 resolves the main issue: ensure new links are valid.
This does not protect against link rot, though. We might want a separate CI job for that. Perhaps that one should operate on the generated site as an E2E test.

[straight-shoota]

I believe we should also remove the dead links plugin (added in #637).
It automatically converts links marked as dead with a link to the web archive. But it's not fully automatic, we still need to manually mark links as dead. And it's not fail-safe because links might not exist in the archive, or contain unrelated content.
Instead, we should replace dead links directly with archive links in the sources.

[HertzDevil]

Somewhat related, we might want to do the same in the main Crystal repository to catch dead links in the standard library docs, to take care of things like crystal-lang/crystal#15800