Tradeoffs of specutils subclassing and prospects for assessing alternatives

[gully]

I'm on a chat with @SujayShankar2001 right now and we're talking about tradeoffs with specutils subclassing.

See related issue in muler: https://github.com/OttoStruve/muler/issues/149

In our current gollum/muler ecosystem, here are the functionalities that "touch" specutils:

• muler-- uncertainties
• gollum-- rv shift
• both astropy quantities for wavelengths and flux units
• math operations
• linear interpolation
• blackbody function

We make lots of use of the hidden function _copy() which we both agree is slow and undesirable.