Hi fjulianom,
Reading through your posts and trying to understand the topology I would not recommend a cluster setup between sites.
You will only benefit from the cluster if all members are L2 connected. If they are on different sites, which are L3 connected you will lose most of the benefits of the cluster, like fast failover and so one.
Another point against the cluster across sites might be, that AP's and clients are balanced across all members. This means, that AP's and clients from one site might be connected to a controller on a different site. From my point of view, this is not what you want and something I would never recommend.
If you will have only one controller per site, you should run them as a single controller managed by the MM. As already said, controller failing is very uncommon.
If you start phase two of the project and you get a second controller per site, I would create a two-node cluster per site. And even with this setup, you can let the AP's failover to a cluster on a different site, for the case even the two controllers on the same site fail, at the same time.
hope this makes sense to you.
BR
Florian