Revisit the use of cluster_public_ip

cluster_public_ip has been introduced to handle the case where cluster_external_ip is not reachable from bootstrap cluster, but hidden behind an openstack floating IP.

As it was required for libvirt-metal, we introduced another way to have a cluster_public_endpoint used to reach cluster from the outside that is different from cluster_external_ip, by adding this public ip to cluster_api_cert_extra_SANs, and derive a "public cluster kubeconfig" pointing to that public endpoint. See sylva-projects/sylva-elements/helm-charts/sylva-capi-cluster!130 (merged) and !791 (merged)

The use of cluster_public_ip for openstack could be revisited to use the same mechanism, and rely on cluster_public_endpoint like libvirt-metal. This way management-cluster-kubeconfig (used by cluster-API) would be bound to cluster VIP instead of openstack floating-IP, which seems more clean & robust IMO.