Link checker selectiveness: run only on changed files on small PRs

[holly-cummins]

This is the promised follow-on to #2685 to reduce the build-time impact. It only checks changed files for internal dead links. This does introduce a gap where if a file is moved, any files that reference that file wouldn't be checked, so build would be green. However, the external link checker (#2697) would pick that up on its scheduled runs and raise a defect. I think the trade-off is worth it for the improved build speed.

If someone has a PR which is lagging way behind main, they'll also get a slow build, but I think we can live with that.

Here's the logic:

• Compares the built _site/ against the gh-pages branch (the deployed site) using rsync to find changed HTML files (using a git diff would pick up adoc changes instead, and we'd have to map from them to the html)
• When ≤15 HTML files changed and no infrastructure files (build.yml, pom.xml, src/test/java/) were modified, runs the link crawler only on the changed pages with a depth-1 check (verifies outbound links from those pages)
• Falls back to a full crawl when >15 files changed, infrastructure changed, or gh-pages comparison fails
• Skips the link crawl entirely when no HTML files changed

In this PR, we can always trigger the 'full' scenario so can only exercise the 'full' paths, but I've tried my best to test the others locally. I'll also keep an eye on PRs that go in after this merges to try to validate the scope detection is working as expected.

Attempts at testing it

• Ran incremental mode against quarkus.io with 2 paths (/blog/, /about/) — crawled 217 pages (depth-1) vs thousands for a full crawl, 0 broken links
• Tested scope determination script locally: Claude claims it correctly detects changed files, skips unchanged files, converts paths to URLs
• Tested no-changes edge case: correctly produces mode=skip

[github-actions]

🙈 The PR is closed and the preview is expired.