Change rancher-save-images.sh and rancher-load-images.sh to use the rancher-images.txt file instead of hard-coded images

Some entries can have a type like inetorgperson or
inetOrgPerson. Those should be treated the same, but our comparison
was case sensitive.

Issue: we don't update version info if stats hasn't changed, causing
k8s version to not reflect on cluster update

This reverts commit d9c16bbe.

from the user cluster. There used to be a bug when the namespaces were
fetched from the management plane

Problem:
An error is being logged but not returned. This causes the caller to
panic when attempting to dereference the pointer

Solution:
Return the error instead of logging it allowing caller to catch the
error

Ensure that closed watchers are always stopped.

This would cause a new watcher to be started without ever cleaning up the old one, which was causing memory leaks.

Bugs:
    (*) https://github.com/rancher/rancher/issues/15593

Details:
    (*) Refactor PowerShell scripts
        - doesn't use Kubelet to register at first
        - stop previous processes forcibly via `Stop-Process -Force`
        - separate Flanneld start logic from `config-cni-flannel`
        - embed CNI configuration and networking components running into
        `start-kubelet`
        - add Cloud Provider supported
    (*) Add Cloud Provider configuration when generating Windows agent deployed plan
    (*) Don't stop PowerShell scripts execution when WebSocket connection cutting in and out actively