🛠️ Retire create-worktree.sh now that SWE always creates its own task-branch worktree explicitly

[ZaxShen]

After #170/#171 fixes, every SWE spawn results in two worktrees when only one is needed:

1. `worktree-agent-` — created by CC's automatic WorktreeCreate event firing `scripts/hooks/create-worktree.sh` (because `agents/swe.md` has `isolation: worktree` in its frontmatter).
2. `fix/` — created by SWE's first command (`git worktree add .claude/worktrees/ ` per the new doctrine in 🐛 SWE creates branches; doctrine says bro owns branch creation #170).

The first one is leftover from the original Layer-2 workaround for CC's bug #27134/#44965 (worktree_add defaulting to `origin/HEAD` instead of `HEAD`). With SWE now always creating its own worktree explicitly via Bash, the auto one is redundant — and the new `cleanup-worktree-on-task-close.sh` only removes the task-branch worktree, leaving the auto one to accumulate.

Proposed

• Remove `isolation: worktree` from `agents/swe.md` frontmatter (or change to a value that doesn't trigger CC's auto-worktree if such a value exists).
• Remove `scripts/hooks/create-worktree.sh` + the WorktreeCreate hooks.json entry.
• Verify with a fresh L5 dogfood that only the bro-created task-branch worktree gets created and cleaned up.

Risks

• CC's bug may still bite if a future doctrine change re-introduces auto-worktree.
• Some flows may rely on the agent isolation that CC's worktree provides for free. Worth checking that the explicit Bash-created worktree provides the same isolation guarantees.

Surfaced by

Local h5 re-run after #172 fixes — observed two worktrees in the project after task close, one cleaned, one orphaned.

[ZaxShen]

Closing — focusing on v0.8.x. Planning lives on GitLab milestones (v0.8.0 / v0.8.1). Will re-mirror active items to GH once tri-source schema sync (#48) lands.