test-agent: handle descheduling and rescheduling

If the pod running a test_agent is descheduled and rescheduled, or otherwise fails and starts again, the current interface does not provide and way to handle that. Put differently, the initial implementation in #62 assumes that the pod will run to completion. Robustness and failure recovery need to be added to the traits/framework.

Start by reviewing the initial test_agent implementation in issue #62 and the traits/framework it introduced. Trace how a pod that is descheduled, rescheduled, or restarted is represented and identify the existing recovery boundaries. Done means the framework has a defined way to handle those failures rather than assuming the pod runs to completion.

Written by the indexing model from the issue text.