Move component methods into Fitter, refactor tabulate

[jdtsmith]

Model.tabulate is slow, and relies on Model-recreation and a single hard-coded Fitter class. This is also part of why Model.plot is so slow.

How about the following idea for component computation:

• Elevate Fitter to be a proper super-class.
• Give Fitter it its own component methods, one for each component (starlight, blackbody, modified_blackbody, power_gaussian, power_drude). They will take wavelength and standardized component parameters in a fixed order
• Make Model.tabulate use its own Fitter object compute (creating one if it needs to).
• Model.tabulate can use rich keywords to return a dictionary of data, similar to the IDL pahfit_components, which had:

                          DUST_CONTINUUM=dust_continuum, $
                          TOTAL_DUST_CONTINUUM=dc_tot,STARLIGHT=stars, $
                          TOTAL_CONTINUUM=tot_cont, $
                          DUST_FEATURES=dust_features, $
                          TOTAL_DUST_FEATURES=df_tot, $
                          LINES=lines,TOTAL_LINES=lines_tot, $
                          EXTINCTION_FAC=ext, $
                          EXTINCTION_CURVE=ext_curve, $
                          DISABLE_EXTINCTION=de
  

- We introduce tests to ensure all `Fitter`'s (including the parent class itself) give the same component outfit for the same input.

I think to start the base `Fitter` components would just be lifted from `APFitter`.  

One minor variation of this is to create _two_ abstractions: `Fitter` and `Components`, with the latter doing the component work.  But I think that's likely overkill.