I think you are right, the issue here is that the FeatureTable[Frequency] IDs are the collapsed taxonomy strings, while the FeatureData[Taxonomy] IDs are still the hashed sequence values. We don't have a method in QIIME 2 currently to collapse the FeatureData[Taxonomy], so you will have to do that manually (I would write some kind of script for that, personally).
Taking a step back, why can't you use the uncollapsed table here?
Thanks!