Optimize Sub and Add backward kernels

Required prerequisites

As pointed out in !138 (comment 3182436) and !137 (comment 3182429), these kernels could be optimized.

added EnhancementPerformance ⚡️ PriorityLow StatusTODO TopicOperator labels

assigned to @jeromeh

See optimizations for fw kernels here : !40 (merged)