Fornax_2019 angle-dependent fluxes no longer function

[sybenzvi]

I discovered several problems with the Fornax_2019 class, which includes angle-dependent fluxes.

1. The class caches its calculation of the angle-dependent flux, based on a HEALPix map, using an option cache_flux in the constructor. The default value is False and the class decorator does not seem to allow overriding the default value.
2. The get_flux member inherited from SupernovaModel cannot handle angle dependence.
3. The innards of the class need to completely upgraded to handle the FluxContainer added to snewpy. Additional modernization may be needed.

This class needs a lot of TLC to be brought up to snuff. Do we remove it?

[JostMigenda]

1. The class caches its calculation of the angle-dependent flux, based on a HEALPix map, using an option cache_flux in the constructor. The default value is False and the class decorator does not seem to allow overriding the default value.

This is solely a performance optimisation, right? If so, I think it’s not critical to keep this, if this is nontrivial to implement with the decorator. (And who knows, maybe rewriting with the FluxContainer would have some performance benefits anyway, so the need for caching decreases?)

2. The get_flux member inherited from SupernovaModel cannot handle angle dependence.

See the related discussion in #165. I’m not 100% happy with the solution I suggested there (moving the angle dependence into the __init__()), but it should be pretty straightforward to get that to work.

3. The innards of the class need to completely upgraded to handle the FluxContainer added to snewpy. Additional modernization may be needed.

This class needs a lot of TLC to be brought up to snuff. Do we remove it?

Hm … in principle, it would be good to support different observer directions in SNEWPY; there are some interesting analyses that could be done with this. I notice that the Fornax_2024 models also have directional information, though it’s not implemented in #375 yet. Have you had a look at the format of those? Would we be able to reuse much of the implementation between the 2024 and 2019 models?

[sybenzvi]

@JostMigenda: you are correct, the FITS cache is solely a performance optimization. We can figure out a better way... or drop it.

The Fornax_2024 models are available in ASCII files from the Burrows group, not HDF5, so I'll need to repack the data to recycle the code used in Fornax_2019, or come up with a better internal representation.

[JostMigenda]

See also #223 and #165.