The new sparse implementation is awesome! There's a bunch of heritage code that can probably be scrapped as we move towards supporting just the sparse implementation. In particular the multiorder module is largely deprecated but has a few useful parts: Chebyshev coefficients and some hard-coded attributes, such as speed of light, n_pix, etc.

We should completely remove read_native_PHOENIX_model now that we've migrated to gollum-based inputs.