Fix inconsistent state after error

[jankatins]

There are two places where we could end up with inconsistent state between what the DB has and what the incremental copy state/ the file dependency hashes know about:

• Incremental copy: If one would do a TRUNCATE/rebuild of the target table and the resulting full load would error, the next run would be an incremental one (as the old comparison value would still be there) despite the table not fully loaded
• file dependencies: if you do a deploy, which results in a rerun and this rerun fails + this deploy is then reverted, the old file dependencies would match and not trigger a rerun despite the DB not being in a fully functional state

In both cases, we now simply delete the old state before the action (copy/execute) is attempted. This should trigger a full copy in the next run in case the copy is not successful/ a full run if the deploy is reverted.

Closes: #18

[jankatins]

I tested this by triggering the problems locally and this fixes the problems. I will report back in a few days and a few more regular loads...

[jankatins]

I've tested this now locally for about a week worth of loads (most if not all of them not triggering the new codepaths itself, as the loads were without problems). In artificial runs which simulated the error conditions it worked as intended. So I feel comfortable to merge this.