Joint/Multi-Fit in the Model<->Fitter paradigm

[jdtsmith]

Let's discuss here the logic of how multi-segment fitting can/will go, now that we have abstracted out Fitter and its spoon-feeding parent Model.

[jdtsmith]

• A list of Spectrum1Ds becomes a valid fittable.
• A jointly fitted feature (line, practically) can apply to more than one segment. Model decides which segments, just as it decides which lines to "cull" (i.e. no segments).
• If a list of spectra is given for fit, Model first checks if the fitter has the JointFit capability, raises PAHFITCapabilityError if not.
• If so, Fitter.fit (and others) will have to accept a list of numpy arrays for its wavelength argument.
• Model informs Fitter of the joint fit via a special keyword to add_feature_line (etc., should we choose to do this for other feature kinds) — e.g. segments — which provides a tuple of segment indices (often, just one) to which this feature applies.
• In this scenario, if the user has hard-coded line fwhm into the table (or tied it to another parameter), an error is thrown — a joint fit makes no sense except when the instrument pack is controlling.
• Fitter does what it needs to do to arrange for a joint fit for that feature. When it is queried for the fit results for that feature, it must return a single best-fit value for power. The value return for fwhm is ignored (and probably shouldn't actually be returned unless it was actually fit, i.e. not in this scenario).