mm, swap: defer xswap shrink to workqueue to avoid lock recursion

将xswap收缩移到workqueue执行，避免在__free_cluster()持有ci->lock时调用xswap_unmap_clusters()导致锁递归及页表失效后崩溃。