Sharp Capacity Thresholds in Linear Associative Memory: From Winner-Take-All to Listwise Retrieval — ThinkLLM