Internal DNS keeps writing network contexts that have not changed #855

Summary

Two internal DNS controllers write network contexts that have not changed. The hub's project reconciler stamped a context's region and cleared it again in a loop when the context's network had DNS off. #847 removed that loop, and v0.34.2 carries it to production's hub since 2026-11-03. Each cell still patches the hub's status of every internal DNS context it serves every 10 seconds, whether the status changed or not.

What we know

• The loop: the reconcile stamps the region annotation, then disable clears it with the rest of the intent. Each write is an event on the same reconciler, so the context never settled. On 2026-11-03 production's Milo took 434 NetworkContext patches a minute from it, and 0.1 a minute after v0.34.2 rolled.
• The report: report patches the hub's status on every reconcile, and every reconcile requeues after 10 s. The patch renewed a lease when 2d564057 added it. 8a3d3fe8 removed the lease, and the patch stayed.
• When a patch changes nothing, the API server stores nothing. On 2026-11-03 staging's hub took about 248 status patches a minute from its two cells and stored 3.6 of them, so the cost is requests to Karmada.

Acceptance criteria

• A network context whose network has DNS off is written once, and the next reconcile writes nothing: fix: Let a context settle when its network's DNS is off #847, read in production after v0.34.2.
• A cell sends its report only when the status it reports changes.

Non-goals

• Replacing the cell's 10 s poll with events and a deadline. That is a separate change with its own review.