Previously, files are backuped to logseq/bak when logseq detects
there're differences between the db and disk. But it has two problems:
1. Only a few of users know `logseq/bak`, other users think that their
data has been lost.
2. Files in the logseq/bak folder are never truncated.
This PR backups old file in DB with timestamp suffixes instead of
logesq/bak, and only keep the latest 10 versions of any changed file.