Kubernetes.io Docs: Requesting clarification

In the page Options for Highly Available topology - Kubernetes

It is stated that each kube-apiserver communicates with each etcd node. It is also stated in official etcd documentation that it uses RAFT consistency - which only allows for at most a single leader in an etcd cluster at any given time.

Can you clarify/explain how can the control plane nodes communicate with different etcd nodes in either the stacked or external schemes?

Especially the stacked diagram confuses me - even if the etcd instance is in the same node, shouldn’t the apiserver communicate only with the leader, no matter on which node it may be?