Currently if dataset is created in some job, we associate to it (set job_id to it) even if job failed and we never change this association. User can re-run job with leaving that dataset chain in the code and because of the checkpoints we will skip dataset creation (we won't create new version) but job_id of that dataset will not change, i.e. we won't associate it to a new job. This creates couple of issues, one of being that dataset is not visible in second, or any subsequent jobs in UI even though user can clearly see it's chain in the codebase.

Solution is to:

• If we find checkpoint for some dataset, skip it's creation as we do now but update job_id to the current job
• In order to still see this dataset in previous job runs, we need to somehow save those "old" job_ids. This can be done with many-to-many relationship or similar.