Allow a Lost run to be overruled by delayed events #4236
Description
One pattern I notice is that after a Run has been marked Lost by Lightning, often the work has all been done and we were just waiting for a message.
Sometimes after Lost, the worker will reconnect and try to send all messages. And they'll be rejected because the run was Lost.
We should consider engineering robustness into Lightning. A run can be Lost, sure, but if more information comes in after this, it should still listen, and the status should be updated.
This may well need support from the Worker. I think we should discuss next year.