Possibly invert Fourier-transformation responsibilities
Right now, the domain-space computes the data for the codomain-space. The field then uses this data together with the new data. For the spherical spaces this produces strange if-else blocks in the calc_transform methods.
It would be clearer when the codomain-space computed its data directly: e.g. lm_space then has two sub-methods: transform_from_hp and transform_from_gl