Catch Me If You Can: Uncovering Malicious Behavior in Container Images