implement restore for incremental backup
This diff adds the restore for incremental backup, in order not to change the api name, I use a flag in restoreDBFromS3 in admin_handler to distinguish from the previous restore.
It will first check the latest timestamp for the db from the restore request, and download all the files from the latest backup. Then, it will download all the missing files from previous backup corresponding to the map in backup descriptor. Finally, add the db.
There is one problem about lock. For example, if someone is removing the db and it is incrementally backuping the db, it will cause some problem. I did not find a good way to share a lock between db_manager and the backup_manager. Furthermore, it is better to have a fine-grained lock for each db. Otherwise, the system will be inefficient (every operation will acquire the only one lock).