GLiRA: Black-Box Membership Inference Attack via Knowledge Distillation