DistributedLB: avoid leaking memory from tuple reductions