Restructure generation of longitudinal data

Improve coding structure for generating longitudinal outcomes, e.g. building a growth model starting with an initialized outcome, and building later scores through lagged values, demographic, and latent student factor predictors.

This structure should (if possible) be made general to all outcome types (having a linear, continuously-valued latent value at its core) which can be left as a continuous outcome, or given a non-linear transformation ex post.

The issue names no files, tests, or entry points. Start by locating the current code that generates longitudinal outcomes and trace initialized outcomes, lagged values, demographic predictors, and latent student factors. Determine whether a shared latent outcome structure can support continuous values and nonlinear transformations, then identify checks needed to confirm equivalent behavior.

Written by the indexing model from the issue text.