docs(ci): explain nested testsuite workflow refs

[bketelsen]

What problem are you solving?

Stable-promotion triage can mistake test_ref: v1 for proof that every testsuite fix is current. The nested reusable workflow has a separate uses ref, so workflow-level fixes such as VM disk sizing remain stale when that ref is pinned. This documents the distinction found while investigating #929 and provides consumer validation for the owner-side projectbluefin/actions fix.

• I am using an agent and I take responsibility for this PR

Changes

• add the nested workflow/test checkout ref split to the CI failure-mode table
• direct promotion triage to verify the reusable workflow shown in the run log

Testing

• just check
• python3 .github/scripts/validate-docs.py
• pre-commit run --files docs/skills/ci/references/failure-modes.md

Checklist

• Conventional commit message
• No hardcoded secrets or credentials
• Documentation updated for the reusable CI procedure
• No image or release-gate behavior changed

Refs #929

Owner-side fix: projectbluefin/actions#409

[castrojo]

Superseded by #1077, which carries this change re-cut from current main.

This branch could no longer be merged. Measured today:

$ git merge --no-commit --no-ff origin/main
Automatic merge failed; fix conflicts and then commit the result.

$ git diff --name-only --diff-filter=U | wc -l
16

The conflicts are Containerfile, .github/workflows/{bonedigger,lab-check,pr-validation,promote-testing-to-main,run-testsuite}.yml, build_files/base/20-tests.sh, build_files/packages/base.toml, six skill documents, tests/unit/04-install-kernel-akmods_test.bats, and renovate.json (deleted on main, modified here) — for a pull request whose entire contribution is 13 lines in one documentation file.

Resolving those by hand would produce a diff that no reviewer could check against the change it claims to make, and every conflict resolved wrong is a workflow or a Containerfile silently reverted. Re-cutting is both smaller and safer.

The content still stands. docs/skills/ci/references/failure-modes.md has grown substantially on main since this was opened, but nothing in it covers the nested-ref distinction, so #1077 carries the triage row and the explanation, with authorship preserved in the commit trailer.

Closing in favour of #1077.