Why limit mesh topologies to cluster-trees? There are plenty of more efficient ways to route data...
I'd design it having message flooding for peer discovery (no hop limit, but a bandwidth limit - never use more than 0.1% of the total throughput for flooded messages).
Then, once connections are established, pick a few best routes, and send some proportion of data down each. Weight the paths via a cost function that takes into account load of each node, power use/availability of each node, impact of each flow on other flows (prefer getting nodes to transmit who cause least interference to other flows), etc.
Over time, adjust proportions of data down different routes to minimize the cost function.
I'd design it having message flooding for peer discovery (no hop limit, but a bandwidth limit - never use more than 0.1% of the total throughput for flooded messages).
Then, once connections are established, pick a few best routes, and send some proportion of data down each. Weight the paths via a cost function that takes into account load of each node, power use/availability of each node, impact of each flow on other flows (prefer getting nodes to transmit who cause least interference to other flows), etc.
Over time, adjust proportions of data down different routes to minimize the cost function.