Deprecate OneSpec

[kbwestfall]

We should replace OneSpec with pypeit.core.spectrum.Spectrum in the short term and specutils.Spectrum in the long term.

[bpholden]

The OneSpec class is used in the coadd and telluric functions.
A number of the documents for users mentions this class.

The plan is to replace the OneSpec class with the specutils.Spectrum class (not Spectrum1D, that is being depreciated.)

OneSpec attributes

• wave
• wave_grid_mid
• flux
• mask (True if good, False if bad)
• sigma
• inverse variance
• fluxed
• history
• telluric
• object model
• head0
• ext_mode (extraction mode?)

specutils.Spectrum attributes

• wavelength
• flux
• mask (False if good, True if bad) - see specutils.Spectrum.mask
• uncertainty (can be sigma, variance, or inverse variance)
• meta - a dictionary of stuff

There are a few challenges:

1. there is only one wavelength solution in Spectrum, and OneSpec has two
2. you cannot add more than one spectrum, and the OneSpec has the telluric as well as the object model
3. you have only one uncertainty and the OneSpec has two
4. the mask definition has the opposite sign

The official specutils response to this is "put everything in meta."

This yields a straightforward replacement to the OneSpec in the coadd.py construction where the wave_grid_mid , telluric, object model, head0, ext_mode, and history are all placed into elements in the meta attribute.

There already exists a pypeit_onespec_loader in pypeit_loaders.py that does most of this.

This leaves a couple of issues.

1. the mask has the opposite sign, so every routine needs to know this or, we flip the sign when the item is read in - addressing this in pypeit_loaders.py is left as a TODO
2. do we keep both uncertainty maps or only use one? for the current draft the inverse variance is kept but we all know that dividing by small numbers is a bad idea. We could keep both and just have a convetion that one is the uncertainty array and the second is yet another element in the meta
3. A number of checks of the input object use the FITS keyword DMODCLS which is written to the data unit of the FITS file that contains the spectrum. The specutils.Spectrum does not generate such a keyword. Instead it creates a FITS table or WCS spectrum depending on the details.