Schema lookups stop filling the log with false errors

Whenever an agent created or updated a node, the portal checked the content against the node type's schema. To find out what that schema is, it applied the node type's configuration to a throwaway hub — and that hub was given the full machinery a real, long-lived node gets: the compile watcher, the release watcher, the sources watcher, and the compile-state mirror.

None of that had anything to do on a hub that exists for a fraction of a millisecond. Each watcher would open a connection to the mesh and then fault as the hub was disposed out from under it, reporting the teardown as a failure and trying to re-establish. One schema check produced around twenty error and warning lines — none of them describing a real problem, all of them competing for attention with the ones that do.

Throwaway lookup hubs are now built without that machinery. They still carry everything the lookup actually reads, so schema validation and the node type's data-model view behave exactly as before — they just no longer report their own shutdown as a fault.

A rejected write also got cheaper: the error message and the schema the agent needs to correct itself are now read in one pass instead of two.