Update: hexo-generator-feed/index to v4, hexo-fs to v5

[Aetf]

Stacked on #56 — these are Hexo 8-era releases (hexo-generator-feed 4 depends on hexo-util 4, hexo-generator-index 4 on hexo-pagination 3), so they can't go in before the Hexo 8 bump. Retarget to develop once #56 merges.

hexo-generator-feed   3.0.0 -> 4.0.0
hexo-generator-index  3.0.0 -> 4.0.0
hexo-fs               4.1.3 -> 5.0.1

Verification

Diffed a clean build against a clean pre-bump build. Every HTML page is identical once the random prism code-block ids are normalised. Only two outputs change, both from hexo-generator-feed 4's rewritten generator:

atom.xml — elements reordered, and it now emits a feed-level <rights> and a per-entry <author>:

	feed children	entry children
v3	author, entry, generator, id, link, subtitle, title, updated	category, content, id, link, published, summary, title, updated
v4	... + rights	... + author

Parsed both with ElementTree — same 21 entries, same ids, same titles, all still carrying <content>. Strictly richer, nothing dropped.

search.xml — same 21 entries in a different order. Compared per-URL content hashes: identical URL set, identical content per URL, order only. It's a search corpus, so the order carries no meaning.

Post listing order and pagination are unchanged (/blog/, /blog/page/2, /blog/page/3 all match). 43/43 tests pass; headless render of /, /blog, /about, /projects, /research and a post is clean with zero page errors.

Still not updated

katex stays on ^0.16 — see #55 for why 0.18 can't move until NexT's vendored KaTeX CSS does. With this merged that's the only remaining outdated dependency in the tree.

🤖 Generated with Claude Code

[Aetf]

Note: GitHub Actions won't run on this PR. ci-workflow.yml triggers only on push to develop and pull_request into develop, and this one targets chore/hexo-8 so it can be reviewed as a separate change. It'll pick up CI as soon as it's retargeted to develop after #56 merges.

Verified locally in the meantime: clean npm ci on Node 20, npm run build clean, 43/43 tests, plus the build diff and headless render described above.

[Aetf]

Superseded by #60 — same commit rebased directly onto develop.

GitHub auto-closed this when I deleted chore/hexo-8 during post-merge branch cleanup, and it won't let a PR whose base branch was deleted be reopened (I recreated the base branch and retried; reopenPullRequest still refuses). Lesson noted: retarget stacked PRs before deleting the branch they're stacked on.