[GHC] #12908: Tuple constraints refactoring seems to regress performance considerably