v0.4.0

• New models.SpecModel class that handles much of the conversion from FSPS spectra to observed frame spectra (redshifting, smoothing, dimming, spectroscopic calibration, filter projections) internally instead of relying on source classes.
• The SpecModel class enables analytic marginalization of emission line amplitudes, with or without FSPS based priors
• A new mixture model option in the likelihood to handle outlier points
• A noise model kernel for photometric calibration offsets.
• Rename mean_model to predict (old method kept for backwards compatibility)
• Some fixes to priors and optimization
• Python3 compatibility improvements (now developed and tested with Python3)