SaveCallback: test integrity of existing snapshots

When trying to save a snapshot to a file that already has that snapshot we currently use compare in most cluster simulations, i.e. load the snapshot and compare the data. I recently had simulations dying because I ran out of disk quota leaving me with broken snapshots (because they couldn't be written fully).

I would suggest that compare could be improved to wrap the loading in a try catch and just write the new data in case the old snapshot is broken. (JLD2 implements delete!(g, name))