worktree: cover parallel slots and shared state

[bert-script]

Why

Two gaps came up when a consumer asked for several worktrees for parallel work:

• "Set me up three worktrees", with no work named, has no row in the argument table. Slugging the sentence gives one branch named after the request.
• Step 3 isolates what sits inside the checkout. Two live worktrees also share whatever the env files point at: a dev database, a dev-server port, a cache namespace. Nothing fails; the second worktree quietly uses the first one's, and it shows up later as a migration nobody ran.

What

• § 1 gains the N-slots shape: cut N worktrees on placeholder branches (wt1, wt2, …), say they are placeholders, hand back the rename. No invented ticket keys.
• § 3 gains one exception for parallel slots: name the shared resources and the per-worktree arrangement. The specifics belong in the repo's house rules (references/<repo-name>.md in skills/, .claude/project-context.md's ## Worktrees in the plugin), never derived on a hunch.
• Both copies change; each points at its own house-rules home.

From Lokale Keuze's copy of this skill (LK-1305). Their own reference file stays with them.

No plugin.json version bump. This is one of eight parallel PRs from the same consumer, and each bump would collide on the same line. Bump on merge as you prefer.

Test plan

• Both copies carry the same two paragraphs, apart from where house rules live.
• CI green

🤖 Generated with Claude Code

[bjvle]

The rename advice undoes the skill's own safety step (Step 2):

`--unset-upstream` is load-bearing. Branching from `origin/<integration>` makes git track that remote branch, so a later `git push` would update the integration branch itself.
The following --no-track flag prevents it:

`git -C <path> switch -c <REAL-BRANCH> --no-track <base>`