For foreground GC, we cache node blocks in victim section and set them
dirty, then we call sync_node_pages to flush these node pages, but
meanwhile, those node pages which does not locate in victim section
will be flushed together, so more bandwidth and continuous free space
would be occupied.

So for this condition, it's better to leave those unrelated node page
in cache for further write hit, and let CP or VM to flush them afterward.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>