don't rely on docker inspect for determining container death

Docker can take a while to notice & record the death of containers, so
we perform a more direct check instead.

Fixes #522.

weave

@@ -284,7 +284,7 @@ with_container_netns () {
     shift 1
     if ! "$@" >$IP_TMPOUT 2>$IP_TMPERR ; then
         STATUS=1
-        if [ "$(docker inspect --format='{{.State.Pid}}' $CONTAINER)" != "$CONTAINER_PID" ] ; then
+        if ! kill -0 "$CONTAINER_PID" 2>/dev/null ; then
             echo "Container $CONTAINER died" >&2
         else
             echo "Failure during network configuration for container $CONTAINER:" >&2