Add option to modify only input transform for shifts in apply_matrix()?

[rhugonnet]

This is mostly code architecture, it would allow to get rid of redundant apply_rst functions for Affine classes that only perform shifts (NuthKaab, VerticalShift, etc...). And ensure the apply operation is fully consistent across affine classes.