Make sure the telemetry has exactly one reader verifiedimportant

This is the trap that broke one of our own measurements and would break a customer's reporting the same way. AMQP delivers each message to one consumer; a customer's own integration attached to success would take half our telemetry and we would take half of theirs, with neither side seeing an error. The engine's answer is per-message queue names: a job can say where its results go, so a customer feed is a different queue rather than a second reader. The related check is that the queues are being read at all — an unread result queue grows without limit and is the quiet way a node runs out of broker memory.