retry certian failed activations in case of system error

some system errors - like a failure to start a container - are safe to retry since we know no observable side effects could have occurred. we should consider doing this by re-queuing the activation request.

Start by tracing how an activation request handles system errors, especially failures to start a container, and where the request could be re-queued. Done means the project has a defined and tested retry behavior for errors with no observable side effects.

Written by the indexing model from the issue text.