fix(vibin): stop worktree-setup scripts aborting with a silent SIGPIPE

[jmagar]

The failure

worktree-new.sh, worktree-sync.sh, and worktree-rm.sh exit 141 with no
output at all — no error, no partial work, nothing to diagnose. Measured in
~/workspace/labby (35 worktrees):

installed (unfixed) script: 7/8 aborted
fixed script:               0/8 aborted

Root cause

Every script resolved the repo root the same way:

ROOT=$(git worktree list --porcelain | awk '/^worktree /{print $2; exit}')

awk's exit closes the pipe after the first record while git is still writing.
git dies of SIGPIPE, set -o pipefail promotes 141 to the pipeline status, and
set -e aborts the script — even though the assignment already succeeded.

It is invisible in a small repo: git emits short porcelain output in one
write() and never observes the closed reader. The crash needs the stream to
exceed git's stdio buffer so a second write() happens. It is also a race,
not a certainty, so a single-shot test misses it.

The fix

All four sites now match once without exiting, so awk drains git's output:

awk '/^worktree /{if(!s){print $2; s=1}}'

The one remaining early-close (git lfs ls-files | head -1) is safe because it
sits inside [[ ]], where a command substitution's status cannot reach
pipefail. That is now commented so a refactor out of [[ ]] doesn't silently
reintroduce the crash.

Tests

New smoke-test section builds a 25-worktree fixture with long branch names
(8362 B of porcelain — asserted to exceed the buffer, since a smaller fixture
proves nothing) and runs the real scripts in a loop. Against the pre-fix
scripts it catches the abort 4/20 and 7/8 of the time.

tests/smoke.sh — all pass, including the two new checks.