Allow for generating covariates for rolling fits from time-changing covariates

[bquistorff]

Right now we assume in effect estimation that X is the same for all treatments. In reality we might to allow it to be some averaging process of time-dependent X.