[FEATURE REQUEST] - Use simulation parameters as starting points when fitting synthetic spectra

[JohnLaMaster]

Problem: In deep learning research, we deal with large volumes of (very often) simulated data. In order to study the effects of the new methodologies, we generally need to fit the output spectra from our models. To get meaningful statistics, we need hundreds to thousands of samples fitted and analyzed for each model. Fitting enough samples to get 500 useful fits after QC can take about a full day. When testing 10-15 model variants (typical in DL), that's a extremely computationally expensive and super time intensive.

Solution: With synthetic data, we have the simulation parameters. Using those as a starting point for the fitting routines should cut down on the fitting times. At minimum, it would be helpful to provide the metabolite quantities if not also the lorentzian and gaussian lineshape values.

Alternatives: The simulation parameters could be used directly as starting points, hard constraints for setting the proportions in the initial values after the preliminary fit, or as soft constraints during the actual fitting.