wrap_up.py --remove crashes on branch cleanup: Gh.pr_list() has no head= argument, and two call sites pass one nf-jymv ← Beads

Twinned from https://github.com/cfsh/ninefold/issues/731 by tools/beads/import_github.py's reconcile pass.

Filed unclaimed by nf-dev-probe-gate on wrap-up — the first session to reach this path since the call was introduced, as far as I can tell.

What happened

./tools/pr/wrap_up.py --remove did the removals, then crashed on the branch-cleanup step and exited 1:

``` ✓ removed worktree /root/nf/devs/probe-gate-f5 ✓ removed worktree /root/nf/devs/probe-gate ✓ pruned stale worktree registrations ⚠ kept local branch feat/per-platform-baseline — could not ask GitHub whether a PR merged it (TypeError: Gh.pr_list() got an unexpected keyword argument 'head') ```

The mismatch

nflib/sessions.py calls pr_list(..., head=...) at two sites — :979 and :1438 — and nflib/gh.py:171 has no such parameter:

```python def pr_list(self, fields, *, state="open", base=None, author=None, limit=100) ```

The comment at sessions.py:975 explains why head= is wanted, and it is right: *"head= rather than filtering a page of recent PRs: a merged branch's PR can be arbitrarily far back"*. So the intent is sound and the argument was simply never added to Gh.pr_list — gh pr list --head <branch> is the flag it wants.

Why it matters more than a kept branch

• ⚠️ It fails safe but reports failure. The tool kept the branch (correct — it could not verify), but the process exits 1 after having successfully removed both worktrees. A CTO or script reading the exit code sees a failed wrap-up for a session that wrapped cleanly, and #635/#638 exist because board state drifting from reality is expensive.
• :1438 is a second, untried site. Whatever calls that path will hit the same TypeError.
• The branch it declined to delete (feat/per-platform-baseline) *was* merged — #716, merge commit 0511e33 — so the fallback cost is one stale local branch on a box whose worktree is now gone.

Fix

Add head: str | None = None to Gh.pr_list and append --head <branch> when set. Two lines, plus a test — ⚠️ and the test matters here, because both call sites are on paths that only run at wrap-up, which is exactly why this shipped: nothing exercises wrap_up.py --remove end-to-end, and a TypeError in a rarely-run arm is invisible until someone wraps.

Related: #635/#638 (board vs reality), #684 (the wrap-up predicate).