CI job integration-chutney-shadow is flaky (due to TCP FIN and/or EPOLLRDHUP misbehavior in shadow)

The job sometimes fails due to arti-bench getting stuck inside the simulation. shadow flags the failure since the script is still running at the configured simulation end time.

This shadow simulation isn't deterministic in CI, at least partly because our CI containers don't permit disabling ASLR. As a result, this failure can occur unpredictably even at the same commit. It is deterministic when run locally (at least on my system), and can be triggered by running with multiple simulation seeds (passing -s to tests/chutney/integration-e2e-shadow). Which seed causes a failure is likely to vary by system.