fix(census): ignore stale merged-remote-only branch in in_flight detection (#158)

[robercano-ghbot]

Closes #158

Problem

loop-census.sh's in-flight detection matched git branch -a --list "*feat/issue-N-*", which includes stale remotes/origin/... refs. After a PR merged, the local remote-tracking ref in the main checkout could survive (until pruned), so a still-open issue whose partial PR already merged kept matching that stale ref and was reported in_flight forever — wedging advance_ready and triggering false stall/escalation.

Fix (fix #2, minimal, single-file)

In loop-census.sh only: when the matched branch for an issue is REMOTE-ONLY (no local branch, only a remotes/... ref), it is treated as in_flight ONLY IF no MERGED PR exists for that bare head branch (bot-gh pr list --state merged --head <branch>). If a merged PR exists, the stale ref is ignored (branch=none) so the issue neither counts as in_flight nor blocks advance. A local branch match keeps prior behavior unchanged; the extra gh call is made only on the remote-only path and is || true-guarded so a transient failure degrades to prior behavior.

Tests

Extended .claude/scripts/loop-census.test.sh (real git repo + stubbed bot-gh): new issue-44 scenario (remote-only ref with a stubbed merged PR → branch=none, not in_flight); issue-42 retained as the no-merged-PR control (stays in_flight), proving the fix doesn't blanket-drop remote-only refs.

Gates

build / lint / test all green (GATES_FILE=.claude/self/gates.json); loop-census.test.sh 45/45.

Review

correctness (opus) APPROVE, tests (sonnet) APPROVE — consensus=all. Non-blocking note: transient-gh-failure fallback is code-inspection-covered, not fixture-covered.

🤖 Generated with Claude Code

[robercano-ghbot]

fix(census): ignore stale merged-remote-only branch in in_flight detection (#158) (not yet reviewed)