test: repair the extension-tests suite broken by the test/ rename - #135

11.0 is `latest` and 10.16 is `previous`, so 10.15 is two lines out of
support and its tree is 1002 files / 55 MB that nothing links to from the
current documentation. It also costs a `config-docs-in-sync` leg on every
PR and every nightly run, cloning core's 10.15 branch to verify pages
nobody reads.

This follows the "Dropping a version" procedure in README.md. `site.yml`
needs no edit because it globs `content/server/*`; deletion is the only
way to unpublish a server version. Keeping the folder unlisted - the
shape used for ocis 7.3 and 8.0 - would not work here anyway, because
config-docs-in-sync iterates `content/server/*/` independently of
site.yml and would keep checking it.

`PUBLISHED_VERSIONS` in go-redirect.js and the version loop in
go-redirect.test.js are two independent hand-maintained lists, and the
suite fails on either one alone. Verified in both directions by reverting
one change at a time: `PUBLISHED_VERSIONS matches the built server version
segments` catches the first, `a published version segment is preserved for
per-version fidelity` the second. Legacy `go.php?to=` links under
/server/10.15/ now fall back to /server/latest/, which is what that list
exists to do.

The `latest` and `next` alias trees move on their own, and 10.15 appeared
in none of the `latest-*`/`previous-*`/`current-*` attributes, so
global-attributes.yml is untouched. `npm run antora && npm test` publishes
exactly 10.16, 11.0, latest and next under public/server/, 51 pass 0 fail.

The historical `== Changes in 10.15.x` sections in server_release_notes.adoc
stay - that page is an archive reaching back to 10.0.7 and is not tied to
which versions are published.

Two things README.md got wrong about its own procedure, found by following
it: it named `test/go-redirect.test.js`, which moved to `extension-tests/`
in #135 and holds a second list it did not mention, and it did not mention
the `sync/manifest.yml` row or the table in the README itself. Both are
now bookkeeping steps. Its example command names a version that still
exists, so it stays copy-pasteable.

Accepted cost, per step 4 of that procedure: nothing redirects a retired
version tree, so inbound https://doc.owncloud.com/server/10.15/... links
now reach the site 404 page.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

Review follow-up to the version drop. Two more places named a path that
moved to extension-tests/ in #135, and one of them is the paragraph that
tells you how to *add* a server version:

`test/go-redirect.test.js` -> `extension-tests/go-redirect.test.js` in the
"Dev version numbers are provisional" note, plus the part it omitted. The
drift test compares PUBLISHED_VERSIONS against the published trees, so a
removal that misses either hand-maintained list fails. An addition does
not: add '11.1' to PUBLISHED_VERSIONS only and the drift test is happy,
while the version loop in the test file silently never covers the new
segment. The note now says to add it in both places and why.

`test/static-files.test.js` -> `extension-tests/static-files.test.js`,
same class of staleness, in the same blockquote.

The "Dropping a version" example is now `rm -r content/server/<version>`
rather than a concrete number. Naming a version that still exists made it
copy-pasteable, which is exactly the hazard: it is the one command in
these docs meant to be pasted, and pasting it deletes a live, supported
documentation tree.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>