Re-index fails on corrupt bundle

The re-indexing process should be more resilient, log an error and simply continue with the next node. It doesn't seem useful to refuse repository startup in this case.