Sticky runners can make errors opaque #158
Description
Summary
In the process of debugging #156, I came across an error wherein checked_migration would fail, but the internal runner remained sticky similar to as occurs in the successful case. Additional logging revealed this was in part related to the MovementMigrator's inability to restart over a previously allocated workspace. However, this one again underscores the challenges that this stickiness can present.
Note
This may be something that requires both fixes from within this repository and with kestrel. It may also be something that is rooted so deeply in how tokio plus multiprocessing logic in this particular context that it cannot be effectively resolved. However, that this has triggered another issue makes it a good candidate for additional review.