For each pcluster, the total compressed buffers are determined in
advance, yet the number of decompressed buffers actually vary.  Too
many decompressed pages can be recorded if one pcluster is highly
compressed or its pcluster size is large.  That takes extra memory
footprints compared to uncompressed filesystems, especially a lot of
I/O in flight on low-ended devices.

Therefore, similar to inplace I/O, pagevec was introduced to reuse
page cache to store these pointers in the time-sharing way since
these pages are actually unused before decompressing.

In order to make it more flexable, a cleaner bufvec is used to
replace the old pagevec stuffs so that

 - Decompressed offsets can be stored inline, thus it can be used
   for the upcoming feature like compressed data deduplication.
   It's calculated by `page_offset(page) - map->m_la';

 - Towards supporting large folios for compressed inodes since
   our final goal is to completely avoid page->private but use
   folio->private only for all page cache pages.
Acked-by: Chao Yu <chao@kernel.org>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Link: https://lore.kernel.org/r/20220715154203.48093-5-hsiangkao@linux.alibaba.com