v1.19.0 — the proof-recovery pipeline goes automatic

Headline: the proof-recovery pipeline goes automatic — the queued-proof backlog now drains itself. A scheduled queue-dispatcher runs the governor-metered dispatcher every 15 minutes (it was previously only an operator-invoked mode), so re-routed and queued proofs flow into PRs that Gate A re-verifies and auto-merges; a close-superseded sweep retires each stranded direct PR once its goal lands, so the backlog shrinks instead of doubling; and docs/recovery.md (linked from a new README Recovery section) documents the whole submission/recovery machinery and the Gate A capacity backstop (ADR-058). Together with the earlier re-route tool and admission self-heal (v1.18.0 / v1.17.1), this closes the loop on "the ruv sledgehammer fix" (#1904) — stranded proofs recover into the metered queue with no manual PR surgery and no lost work.

Added

• Added tools/repo/close_superseded.py and a scheduled close-superseded workflow that closes a stranded direct prove(...) PR once its goal is proved on main (its queued re-route, or a peer, landed) — and never a PR whose goal is still open. This is the safe, incremental implementation of "close the stranded originals once their re-routes land" for the ruv-sledgehammer cleanup (#1904) and the capacity drain (#1909): as the queue-dispatcher lands re-routed proofs, this sweep retires the superseded originals, so the backlog shrinks instead of doubling. Closing needs only the default GITHUB_TOKEN. Pure helpers are pytest-covered; --author and --dry-run are supported.
• Added docs/recovery.md and a Recovery section to the README documenting how the swarm submits proofs through the governor-metered queue, why a proof can get stranded (the direct-submission cutover + Gate A flood, #1904/#1909), and the self-healing recovery pipeline: the admission re-route offer, reroute_stranded.py (re-package without re-proving), the scheduled queue-dispatcher, and the close-superseded sweep — plus the operator quick-reference and the Gate A capacity backstop (ADR-058).
• Added a scheduled queue-dispatcher workflow that drains the queued-proof backlog automatically. The swarm/agent.sh --dispatch-queue dispatcher (SPEC-007-A) previously ran only when an operator invoked it by hand, so queued/prove/* branches accumulated unbounded (the backlog in #1909). The workflow now runs it every 15 minutes (governor-metered via UNSORRY_MAX_OPEN_PROVE_PRS / UNSORRY_MAX_GATE_A_IN_FLIGHT, ADR-058), re-packaging already-locally-verified queued branches into PRs that Gate A re-verifies from scratch. It authenticates with the admin REFRESH_TOKEN (a GITHUB_TOKEN-opened PR would not trigger Gate A) and degrades to a report-only notice when that secret is unset. Self-throttling: it dispatches only while in-flight Gate A work is below the cap and no-ops when full.