docker: configure restart policy for networking pause container

This PR modifies the configuration of the networking pause contaier to include
the "unless-stopped" restart policy. The pause container should always be
restored into a running state until Nomad itself issues a stop command for the
container.

This is not a _perfect_ fix for #12216 but it should cover the 99% use case -
where a pause container gets accidently stopped / killed for some reason. There
is still a possibility where the pause container and main task container are
stopped and started in the order where the bad behavior persists, but this is
fundamentally unavoidable due to how docker itself abstracts and manages the
underlying network namespace referenced by the containers.

Closes #12216

.changelog/15732.txt

@@ -0,0 +1,3 @@
+```release-note:bug
+docker: configure restart policy for bridge network pause container
+```

drivers/docker/network.go

@@ -130,6 +130,12 @@ func (d *Driver) createSandboxContainerConfig(allocID string, createSpec *driver
 			// Set the network mode to none which creates a network namespace
 			// with only a loopback interface.
 			NetworkMode: "none",
+
+			// Set the restart policy to unless-stopped. The pause container should
+			// never not be running until Nomad issues a stop.
+			//
+			// https://docs.docker.com/engine/reference/run/#restart-policies---restart
+			RestartPolicy: docker.RestartUnlessStopped(),
 		},
 	}, nil
 }