Replace 'polling' to 'watch` strategy to improve performance and reduce cpu usage

[denis-tingaikin]

Currently we're using 'polling' strategy to check diff with cluster properties. Its polling clusterproperties every second to detect diff.
We can improve this apporoach with replacing 'polling' to 'watch'.

TODO

1. Do fetch for list as its doing currently https://github.com/networkservicemesh/cmd-cluster-info-k8s/blob/main/main.go#L148-L161 before main loop.
2. Start watcher in the main loop instead https://github.com/networkservicemesh/cmd-cluster-info-k8s/blob/main/main.go#L148-L161
   (see at aboutapiRESTClient.Get().Resource("clusterproperties").Watch method)
3. Correctly handle watcher expiration (Note: watchers have a time limit for watching, so we should re-start watch on expiration)
4. Test that cluster-info example is still working with changes.