podman: recover from killed conmon side process

When podman containers are created by the resource-agent, the podman
runtime spawns a side process (conmon) to monitor the container and
record the exit status.

If the conmon process dies unexpectedly (e.g. kill -9), the podman
container can still be stopped, even if the cli returns a generic
error.

Try to distinguish this specific failure condition and make the stop
operation resilient; when it happens, just log a warning and finish
the usual stop actions.