FWIW, I came up with a better approach to handle this, and have asked for confirmation of the interpretation on the IDNABIS list. So I think u32_normalize is fine, as you explained.
http://www.alvestrand.no/pipermail/idna-update/2011-May/007099.html /Simon
