Anchors whose upstream object is gone are never collected #721

Summary

The operator deletes an anchor only when it sees the anchor's upstream object go through a finalized deletion. An anchor whose upstream object is already gone stays on the Karmada hub until someone deletes it by hand, with everything it owns.

#320 stopped new anchors stranding on deletion. The rechecks after it shipped found no new stranded Gateway in either estate. But both cleanups were done by hand, and they covered only anchors that own a Gateway. One anchor from before the fix still sits on the production hub, owning nothing. The anchors for HTTPRoutes, HTTPProxies and EndpointSlices have not been read in either estate.

Split out of #472, which closes with its first criterion met. Related to #672, whose deleted proxy left its children in its project.

Acceptance criteria

• An anchor whose upstream object is already gone is collected without anyone running a tool
• The anchors left from before chore: use fork of multicluster-runtime for coordinator fix #320 are gone from both hubs, for every kind, not only Gateways