Concurrent node updates keep their place in the queue

Two updates arriving together for a node could each create a queue. Replacing one queue could remove its subscriber while a caller still held it, leaving that caller waiting for a result.

Concurrent callers now share the same queue. Accepted writes keep it alive until their work settles, and shutdown reports an error to pending callers instead of leaving them waiting.

The investigation and ownership rules distinguish the reproduced defect from the intermittent CI timeout that prompted the investigation.