The design which the documentation refers to is this: BGP North-South Routing for VMware Cloud Foundation Instances with Multiple Availability Zones

In short: You'll have two Edge nodes running in AZ1 which are protected with vSphere HA and in case of a failure would failover to AZ2. They peer with a routing device in AZ1 and AZ2 respectively and have traffic steering enabled so that e- and ingress is in AZ1 primarily.

From network perspective this is a bad design and leads into long service interruption in case AZ1 fails. I would rather expand this cluster with a second pair of Edge nodes, keep them in AZ2 and peer it accordingly without traffic steering enabled. You'll end up having asymmetric routing but advantages of having almost zero failover time because you'll have active components in every AZ running (no need for failing over Edge nodes between AZ).