Stopped nodes in VMSS/MP are not cleaned up

[mweibel]

/kind bug

What steps did you take and what happened:
When for some reason a node is in PowerState: Stopped, the AzureMachinePoolMachine controller does not detect it and the stopped node stays until detected manually in the VMSS.

To reproduce, scale up a MachinePool, SSH into the machine and execute a shutdown.

What did you expect to happen:
Stopped nodes are treated like failed nodes and get removed automatically.

Anything else you would like to add:
PowerStates are only exposed in the instanceView or when using list instances with $expand=instanceView. I wonder if we should switch to either of that API and take the statuses property into account. Example listInstancesExpandInstanceView.json

Environment:

• cluster-api-provider-azure version: latest main
• Kubernetes version: (use kubectl version): 1.32.x
• OS (e.g. from /etc/os-release): windows/linux