Protection against OOM when reading corrupted bincode

[spirali]

No description provided.

[Kobzol]

Thanks! Do you think that we can write a test for this? :D

[spirali]

It is actually quite difficult, it seems that we need to write a large number on position where size of collection is written.
I can craft manually the file, but it would need to be be updated whenever the journal format changes.

I have put on my roadmap migration to bincode 2.0, so I will think about it again when I will do the update, as journal format would also need an update.

[Kobzol]

I think that we could use a custom serialize impl that would write a bogus size and then just small data.

[Kobzol]

Also, when we load 2 (or N) identical events in a row, we should probably report an error (or at least a warning), because it's highly unlikely that several identical events (including identical timestamps) will be stored in the journal.