Tight Bounds for Learning RUMs from Small Slates