mm/mglru: reduce folios pulled from the oldest gen in inc_min_seq()

减小 inc_min_seq() 每次从最老代拉取的批量（原 MAX_LRU_BATCH），缩短持锁时间，让并发 reclaim/lru_cache_drain 等有机会推进。