Functions rarely have a bunch of arguments of the same type, so it’s not all that useful. Contrast it with, for example, decoding, where an on-disk structure can be an array of ten 32-bit integers where each one means a different thing.
I think an extensible record solution similar to vinyl wouldn’t need any additional abstractions whatsoever, just a properly lazy type family arrangement, which GHC currently does not have. This would then allow records to use dictionaries, lists and arrays (both mutable and immutable) on the data level, which would ultimately solve all of the record woes, not just the aforementioned tip of the iceberg notation issue.
With something like that in place the question of whether Haskell’s ADTs really need named fields would become quite a bit more interesting.