Depending on the database, changing the storage layer might require changing almost everything: the query planner, optimizer and executioner intrinsically need to know how storage works to make queries efficient.

Aurora most definitely does not do that for Postgres. Postgres’s storage layer is separate enough that this is quite possible.

Edit: maybe the confusion here is about what “storage layer” means here. The physical row encoding is still the same. What’s different in Aurora (there are white papers on this) is how the WAL records get applied directly to encoding aware distributed as opposed to separately syncing the heap storage to disk.