crucible agent's apply_smf() is not removing all instances

Something is not cleaning up smf instances when regions are deleted.

After running a test that creates then deletes a disk in a loop, I can see both the
memory usage going up for agent, and the log file listing more and more messages each time apply_smf() runs:

Message like these:

Nov 29 06:27:36.485 ERRO remove downstairs instance: downstairs-66dab510-7749-4191-adb3-330aec2124d7, component: worker
Nov 29 06:27:36.485 INFO instance states: (Some(Disabled), None), component: worker
Nov 29 06:27:36.485 ERRO remove downstairs instance: downstairs-98be2db0-165b-4adb-9c3d-52f2afab8fed, component: worker
Nov 29 06:27:36.486 INFO instance states: (Some(Disabled), None), component: worker
Nov 29 06:27:36.486 ERRO remove downstairs instance: downstairs-0c941299-6abe-427e-9d39-c9ff67f2aa6b, component: worker
Nov 29 06:27:36.487 INFO instance states: (Some(Disabled), None), component: worker
Nov 29 06:27:36.487 ERRO remove downstairs instance: downstairs-de23a0eb-1c79-4f28-8eb5-df689ad5dbd7, component: worker
Nov 29 06:27:36.487 INFO instance states: (Some(Disabled), None), component: worker

Each loop of the create/delete disk script results in another pair of messages printed.

Something yet TBD is not completely removing items from the svc.instances() list.

Start at the apply_smf() entry point and inspect how svc.instances() is updated when regions are deleted. Reproduce the create/delete disk loop described in the issue, then trace why disabled instances remain and why each run adds messages. Done means deleted instances are removed and repeated runs no longer increase memory use or duplicate removal logs.