Refactor forecasts and draft plots

[swo]

• When predicting, do not distinguish between "forecast" and "postcheck." Instead, make predictions for every date & group you have observations. This makes the "scaffold" concept a lot simpler in terms of implementation.
• When doing analyses, we distinguish between prospective "forecast" and retrospective "fit" scores.