As we discovered in #81, sometimes there's an error in how the csv assembly script puts together the full csv data model. If we don't push the commit with the changes to the full csv before the schema convert step, we can't see the diff on the full csv to help with troubleshooting if the conversion step fails.
Fix:
- add an additional push step OR
- find a way to have the existing post-convert push step succeed with changes from the first commit even if the conversion step fails
First one is easier but redundant, second one is nicer but I will need to read up on it.