Using flux-calibrated models instead of normalized models.

The problem: If we implement the mixture model feature, we will probably need to use flux-calibrated model spectra; in other words, the non-normalized spectra natively produced by the model grid (i.e. Phoenix provides real flux units for its models). The reason for needing to use flux-calibrated models should be clear- the relative flux of the mixture model components should scale with different guesses for the effective temperature. There are a few problems that could arise. First, we'll need to toggle on-and-off the normalization whether you're in mixture model or not (that's easy enough). Second, the PCA procedure in the spectral emulation step might acquire more--or at least different--eigenspectra, since the dominant variance will now come from the absolute flux level and not the spectral lines. (I haven't fully fleshed this out, but I suspect the default of applying normalization is there for a reason.) Lastly, the logOmega term might take on a different meaning, or at least different absolute values.

Suggested solution
Just experiment with it and see how it performs. :)

Start by locating the current spectral normalization and PCA emulation workflow, then compare its behavior with flux-calibrated model spectra as described in the issue. Investigate how mixture-model fitting changes the eigenspectra and logOmega term; done requires documented experimental results and a justified normalization approach.