Handling chng "missingness"

- if we're trying to justify including change at all, having the "ideal" version (without contract induced data gaps) of the signal as a different covariate would be useful. This could be a different data target called e.g. `chng_ideal`. This means rewriting the history using historic lag patterns, but with "future" data (which is only latent because of logistics/contracts/etc).
- imputation/extrapolation? maybe dangerous

The issue names no files, tests, or entry points; start by locating the existing chng signal, its historical lag data, and the forecaster evaluation path. Clarify whether a separate chng_ideal target is wanted and whether imputation or extrapolation is acceptable. Done should include an agreed treatment of contract-induced gaps and validation of the resulting covariate.

Written by the indexing model from the issue text.