Model correction

[MattWellie]

Fixes

• A couple more changes relating to THE COHORT THAT MUST NOT BE NAMED
• That project's lack of a -test bucket means no prior data of any form, which has resulted in exercising a few config paths that weren't exposed before

Proposed Changes

• When running a (now vs. 12 months ago) assessment for new genes, correct the expected data type for the relevant model
• If there's no historic data location, don't try to save any (tries to open the GCP path None/<file>

Checklist

• Linting checks pass