fix(ci): stop skipping conflicts, and keep branches current

[catomean]

Ported from bitbaum/orangecat#645, where these two gaps stranded a PR for
hours and a third was found while auditing the fleet.

The merge train optimised for merging and had nothing to say about not
merging. Three different stuck states all rendered as a polite skip:

• A conflicted PR was treated exactly like "not ready yet" and passed over
  in silence, every sweep, forever. Nothing else will unstick it, so silence is
  the wrong answer. Now it is logged to stderr and written to the job summary.
• Nothing kept branches current, so a PR was merged against a base it may
  not have been proven against, and conflicts surfaced hours after the merge
  that caused them. Now a BEHIND branch is updated first — one per sweep, for
  the same reason only one PR is merged per sweep — which also makes conflicts
  appear minutes after the merge that caused them.
• "No checks reported yet" is transient for a PR opened seconds ago and
  PERMANENT for an old one: GitHub does not retroactively run workflows on a PR
  nobody has pushed to. Over-2h-old checkless PRs are now surfaced in the job
  summary instead of waiting patiently forever.

Behaviour verified against the real script with a fake gh on PATH — conflict
reported and not merged; behind-updated and not merged; clean still merges
(no regression); old checkless PR surfaced; brand-new one stays silent.

🤖 Generated with Claude Code

[catomean]

Closing — this PR is fully superseded, not just conflicting.

PR #310 (merged 2026-08-16, after this PR was opened) deleted the local scripts/ci/auto-merge-sweep.sh and switched this repo's workflow to call the canonical sweep script in maonakamoto/dotfiles instead of carrying a local copy.

I diffed this PR's three intended fixes (loud CONFLICTING reporting, BEHIND branch auto-update, "no checks reported yet >2h" summary warning) against the canonical maonakamoto/dotfiles script that main now calls — they are byte-for-byte already present there (lines 307-410). The rebase conflict is on a file that no longer exists in this repo's tree.

There's nothing left to port; the fix already shipped fleet-wide via the canonical script. Closing rather than resolving a conflict against a deleted target.