0425 — One system manager row; the daemon is the same manager at deployment scope
Edit this pageContext. With consumption loops as worker rows, the daemon (vulkan manager run) needed a claim anchor and a deployment-wide scope. Per-topic manager rows existed from earlier work.
Decision. pkg/systemmanager is the top-level door: the daemon is the SAME manager, claiming the single system manager row instead of a group’s rows. “Run everything in the deployment” is one extra WHERE clause in listWorkers — only a system owner has both topic and group unset. Topic manager rows were deleted.
Consequences. N daemons arbitrate over the system manager row with the same worker_instance machinery as everything else, and the row doubles as the deployment-wide suspend switch. Rejected: per-topic manager rows — nothing runs at topic scope, so their rows were deleted rather than kept as a third tier.