I disagree. Pagination only breaks apart a large result set, which can have any order you want as long as you keep track of what you’ve sent and what you haven’t.
If an API consumer accumulate the items from first page, follows the cursor, accumulates the items from the next page, and continues doing that until there is no more items, they are still receiving the full result set in the original order (which could be totally random as far as the consumer is concerned).
Maybe Twitch is now sorting these by some public or internal id, but that doesn’t mean they have to expose it. From an API provider perspective it might even be easier to implement it that way because an id is much more stable than viewer counts (whereas if viewer counts change between requesting two pages you might have duplicate items).