Working through this ... what is the rationale for having changed the API of logicalrep_read_update? It seems kind of random, especially since no comparable change was made to logicalrep_read_insert. If there's actually a good reason, it seems like it'd apply to both. If there's not, I'd be inclined to not change the API, because this sort of thing is a recipe for bugs when making cross-version patches.
regards, tom lane