handle exceptions in background_jobs properly

[Panaetius]

We perform various migrations in background jobs, but if one step has an unhandled exception, none of the following steps will run. This has caused issues in the past.

At the same time, we want to know if there is something wrong, so we want the k8s job to fail if there is an issue so we know there's something wrong.

To this end, we should catch and collect exceptions in all migration steps and continue on errors, and then raise an exception at the end of the whole job if there were any exceptions.