Fills optional blocks with correctly-shaped empties, filters observations to the model's year range, extends catch to the projection years, and re-keys any survey-index covariance matrices. Whether a missing block is actually a problem is left to the fit-time validation, which knows the model configuration.