What is the best practice building a high bandwith CSV LUN for Hyper-V?

You seem to be caught between a rock and a hard place.  The important factors here are that you seem to be limited by the throughput of a single LUN, and you are dealing with Windows Failover Clustering (which is required for the CSV functionality).

If a single LUN is not possible, then you are going to have an issue with somehow creating a single "volume" at the Windows host.  Remember that Dynamic Volumes (at least the form that comes with Windows itself) are not supported for a Cluster configuration.  You would have to look to a 3rd party for a supportable solution.

But I wonder if there really is a need for a single CSV.  Remember that CSVs are effectively Mount Points that are directly mounted on all nodes at the same time.  You are not limited to a single CSV .. you can have multiple. So the greater question that I have, is what is the limitation to having a single CSV?  Certainly the name space will be different.  All CSVs will be mounted under "C:\ClusterStorage" (where C: is your system volume).  But you could have something like C:\ClusterStorage\CSV1, C:\ClusterStorage\CSV2 .. etc.  Clearly a single VHD could not exist *across* the mount points, but a Virtual Machine could have VHDs mounted from *different* CSVs.

Is there an opportunity to distribute the VHDs across the CSVs, and thereby load balance the I/O requirements?  So it's then becomes less about "How do I make a single LUN perform more I/O than it can", and more about "How can I distribute the workload across multiple LUNs that are CSVs"?  The latter is much more achievable.