Continuous migration is not able to properly store errors

[kaplun]

Expected Behavior

Everytime a MARCXML from legacy is not successfully converted and stored on Labs the content should be stored into inspire_prod_records.

Current Behavior

If there is DB error in the middle, the transaction is not in a clean state, and therefore also storing the failed input in inspire_prod_records is currently failing.

Steps to Reproduce (for bugs)

See: https://sentry.cern.ch/inspire-sentry/inspire-labs/group/822737/

Possible solution

In case of SQL error, we should obtain a clean new session before storing the failed marcxml.