The C in CRDT is for conflict-free, i.e. there must not be any conflicts between operations. Incrementing and decrementing a counter, the final result will be the same independent of the order of operations. You can almost not get any further away from a CRDT than last write wins. And for more complicated merge strategies you are essentially guessing what the desired result is but you do not know and everything really depends on the quality of your guesses.
Like the example with the changed fruits in the list, maybe one or the other should end up in the final result, maybe both, maybe none. The correct conflict resolution strategy is not obvious and highly dependent on the use case, that is why I said that implementations are usually not general or they are only general in the sense that you can plug in any behavior you like. Compare this with a real CRDT like a counter, there there is absolutely no doubt about the desired outcome.
Like the example with the changed fruits in the list, maybe one or the other should end up in the final result, maybe both, maybe none. The correct conflict resolution strategy is not obvious and highly dependent on the use case, that is why I said that implementations are usually not general or they are only general in the sense that you can plug in any behavior you like. Compare this with a real CRDT like a counter, there there is absolutely no doubt about the desired outcome.