Describe how you would implement horizontal pod autoscaling based on custom application metrics rather than standard CPU or memory utilization.
