what’s the right way to upgrade an EKS MangedNodeGroup to a new instance type? last time I changed it, it looked like the replace process killed the old MNG before bringing up the new one, leading to downtime in our cluster