Deal more elegantly with failures

[PGijsbers]

There are several types of failures (e.g., connection timeouts) which are ignored right now, these should be handled appropriately. E.g., just skipping that asset and logging it so it can be synced later, or depending on the error perhaps adding self-imposed wait times.

[PGijsbers]

Some improvements with f71f45e.
I think the preferable next steps would be:

• make it possible to easily re-index failed datasets
• perhaps use the response codes from "normal failures" and detect if there is an unusually high amount of those which may also indicate the connector may want to stop