[Lopu] Skip superseded queued snapshots before expensive work

[lopugit]

What changed

• revalidate the exact PR head/base snapshot only after a worker enters the serialized Lopu agent slot
• treat moved, closed, opted-out, or rebase-owned queued work as a successful bounded no-op
• skip the start comment, checkout, AI, Graphify, and publication for superseded snapshots
• preserve deliberate ai-rebase-paused recovery and all exact-lease publication checks

Why

Live activation after #403 proved the durable queue is non-cancelling, but an older PR #288 worker waited for the slot, then spent 8 minutes on full checkout and Graphify before the final lease correctly rejected a moved base. This moves the same safety decision ahead of all expensive work. Existing queued runs remain untouched; new controller runs gain the early gate after this PR merges.

Evidence: https://github.com/lopugit/thingtime/actions/runs/32822938370

Validation

• node .github/scripts/resolve-pr-conflicts-routing-contract.mjs --self-test
• node .github/scripts/workflow-control-plane-contract.mjs --self-test
• node .github/scripts/promotion-worker-routing-contract.mjs
• all workflow YAML parsed with PyYAML
• git diff --check
• Graphify code graph refreshed and JSON outputs validated