CI: node_logs dumping does not work if mgmt cluster is not capm3-virt

I noticed this in CI run of a platform using a CAPO workload cluster and a CAPM3 workload cluster:

> In namespace 'wc-41769360-rke2-hybrid'
> Cluster= cluster.cluster.x-k8s.io/wc-41769360-rke2-hybrid in namespace wc-41769360-rke2-hybrid
> Machine = machine.cluster.x-k8s.io/wc-41769360-rke2-hybrid-control-plane-5r8nq
>> machine_ip=172.19.xxx.yy
>> Machine IP = 172.19.xxx.yy
error: stat bootstrap-cluster-kubeconfig: no such file or directory
error: stat bootstrap-cluster-kubeconfig: no such file or directory
.gitlab/ci/scripts/dump_machine_logs.sh: line 136: bootstrap_cluster_ip: unbound variable

It seems that the node logs CI script assumes that if we're on capm3 then it means capm3-virt and tries to reach the node from the bootstrap environment kind networking.

/cc @mederic.deverdilhac @loic.nicolle