Archived RDI 1.19.1 pages link to /content/...md paths that 404

[EliShteinman]

Problem

Internal links in the archived RDI 1.19.1 docs on redis.io are broken. They render as literal href="/content/..." paths, which 404.

For example, https://redis.io/docs/latest/integrate/redis-data-integration/1.19.1/installation/install-vm/ has 13 of them, including:

href="/content/integrate/redis-data-integration/1.19.1/installation/install-k8s.md"

That URL returns 404. The page it means, /docs/latest/integrate/redis-data-integration/1.19.1/installation/install-k8s/, returns 200.

A sample of 10 pages under /1.19.1/ has 77 broken links. The archive has 335 such links in total.

Cause

DOC-7086 converted content/integrate/redis-data-integration/1.19.1/ from relref shortcodes to render-hook Markdown links: [text](/content/integrate/redis-data-integration/1.19.1/<page>.md).

Each versioned build job in .github/workflows/main.yml does the following:

1. Strips the version from the archive's self-links with awk. It matches only the relref form: ({{< relref "/<product>/<version>.
2. Runs rsync -a --delete-after to copy the version's content up into the product directory. This also deletes the version directory itself.
3. Runs Hugo.

The Markdown form keeps its /<version>/ segment, and the page it points at no longer exists. render-link.html can't resolve it, so it emits the destination unchanged with a warnf. The build still passes.

The href-diff checks for DOC-7086 compared two ordinary builds. In an ordinary build the version directory still exists, so the links resolve there. The link checks proposed in #3866 and #3867 cover the source and the latest build, not the output of the versioned jobs.

The same awk exists in build_kubernetes, build_rs, and build_redisvl. Their archives have no Markdown self-links yet. They'll break the same way once they're converted.

Fix

A second gsub in each of the four jobs rewrites ](/content/<product>/<version>/ to ](/content/<product>/, the same way the existing one handles relref. See the linked pull request.

A cheap guard could also help: fail a versioned job if its output contains href="/content/. No published page lives under /content/, so any such href is a 404.