sled expungement must clean up failed instances

[davepacheco]

Depends on #4872. Once we have #4872, we should use that mechanism to deal with any instances on sleds that have been expunged. (I think #4872 is not happening all that soon, so in the meantime, instances running on sleds that get expunged will be stuck similar to what happens today if a sled reboots -- see #3633.)

I'm assuming this will clean up both vmm and sled_resource rows for the sled. If not, we'll need to do that separately.