docs(minicloud): correct the merge runbook against what the merges did

[fruch]

What

Corrects docs/plans/minicloud-stack-merge-runbook.md, which landed with #15617 and turned out to be wrong in three places once it was actually executed. Docs only, one file.

1. A GitHub stack object is separate from the base refs, and it is what blocks the merge

The runbook assumed base refs were the whole story, so it said GitHub would auto-retarget on merge and nothing else was needed. In reality these three PRs were bound into stack #15619 (size 3), and while it existed:

• feature(minicloud): run SCT tests against local QEMU/KVM cloud emulation #15617's merge button stayed unavailable even though its base was already master and every required check passed;
• retargeting ci(minicloud): run the minicloud jobs on an sct-runner #15618 was refused outright — Cannot change the base branch because the pull request is part of a stack.

gh stack unstack 15619 dissolved it and left every base ref untouched, so the dependents kept their clean 8- and 2-commit diffs. Now documented with how to read the stack number, that gh stack link is the inverse, and that gh pr merge --rebase --admin gets an admin past a missing approval (enforce_admins is false on master).

2. jenkins/precommit does not necessarily clear when the base becomes master

The runbook claimed it "should flip from fail to pass" after the restack, treating the failure as purely a stacked-PR artifact. It did not flip — the artifact was masking a real failure, and both look identical from outside. #15618's console had them one after the other:

fatal: ambiguous argument 'origin/master..HEAD': unknown revision or path not in the working tree.
ruff-format..............................................................Failed

The first line is the stacking artifact (baseHash derived from an unreachable merge ref). The second was genuine: two files in #15618's own commits failed ruff format --check while passing at master. Now says to check the changed files directly, with the commands, and to fix via --fixup + --autosquash rather than leaving fixup commits.

3. git checkout -B <pr-branch> can silently rebase the wrong branch

git checkout -B feature/minicloud-pipelines pr/15618 fails with "already used by worktree at ..." when that branch is checked out elsewhere — and because the following git rebase is a separate command, it then runs on whatever branch you were already on. Mine produced 10 commits instead of 8 before I caught it. The runbook now uses a scratch branch name and pushes with an explicit HEAD:<branch> refspec.

Also refreshed: Step 1 and 2 marked done with the SHAs the merge actually produced, the stale OLD_BASE=139b5499db replaced (15617's head moved when the runbook commit was added to it — the point of reading it rather than remembering it), --force-with-lease pins switched to git ls-remote, and a note that #15668 must be restacked after any force-push to #15618, not only after its merge.

Tests

Docs only — no code paths touched. pre-commit clean on the changed file.

🤖 Generated with Claude Code

[scylladb-promoter]

✅ Test Summary: PASSED

✅ Precommit: PASSED

Total	Passed	Failed	Skipped
15	15	0	0

✅ Tests: PASSED

Total	Passed	Failed	Errors	Skipped
4656	4625	0	0	31

Full build log