Associativity determines whether a reducing operation can preserve its sequential meaning when work is split across parallel chunks. Operations whose grouping changes state or order are not safely parallelizable without redesigned semantics. section 7.1.2
Sequential versus chunked dropping
Dropping ten elements sequentially produces a stable sum: `(->> (vec (range 1600)) (partition 200) (mapcat #(drop 10 %)) (reduce +))` returns `1222840`. A stateful parallel drop can produce varying results such as `1279155` or `1271155`, because each chunk applies state independently.
In `clojure.core.reducers`, foldable transformations such as mapping, filtering, and concatenation can participate in parallel reduction, while operations such as `take`, `take-while`, and `drop` are marked non-foldable. section 7.1.2
