Durability issue with on-disk storage mode

[DavIvek]

Memgraph version
2.15

Describe the bug
Wrong results when while matching all nodes in database, using on-disk storage mode.

NOTE: This data existed in database before restart

[DavIvek]

The problem is that Memgraph saves the current timestamp only in DiskStorage destructor, but if process gets killed with SIGKILL signal, for example, then destructor won't be called and the timestamp won't be saved.
Because of that, timestamp will start from 1 and old keys will be used, hence user will have to query the database until he gets to the timestamp he used before database was restarted.
To fix this, we should either save every commit timestamp after the transaction finishes, or we could add a flag that will save the timestamp after n commits.

[DavIvek]

An additional issue related to this one, the --data-recovery-on-startup flag doesn't do anything in on-disk storage mode. By default, Memgraph is treating it as always true. We should either handle the case where it is set to false (by deleting rocksdb data dirs) or we should write somewhere in docs, that if the user doesn't want to recover data on startup, he should delete rocksdb data dirs manually.