DisCEdge: Distributed Context Management for Large Language Models at the Edge