Added functionality to build models with two compositions (baseline and follow-up)

[jhmigueles]

Hi @tystan, first congrats for a nice and useful package. I work in several projects in which we measured the movement behaviors twice (baseline and follow-up). We are often interested in running longitudinal analyses to investigate the prospective association of baseline behaviors with outcomes at follow-up, and we use the behaviors measured at follow-up to adjust the models. I added this functionality to the predict_delta_comps function, so that if comps_fup is provided and analysis_type == "longitudinal", then it would calculate the ilr coordinates for the follow-up composition and use them to adjust the linear regression model.

I'm doing this pull request just in case you are interested in adding this to your package, the original functionality is not compromised.