Uncertainty propagation in model evaluation

Evaluated models that contain uncertainties in the parameters (e.g. after having been fit to data), do not provide associated uncertainties in the outputted data array. Related to #474.

For example, consider the following Gaussian:


With uncertainties:

`amplitude = 0.8994641802619674+/-0.1070127529446945`
`mean = 49.698139254800864+/-0.7806634058822939`
`stddev = 5.68318118856827+/-0.7806634058822942`

We'd expect something like this upon evaluating the model:


This is a deep astropy-related issue since the evaluation methods within the models are not aware of the uncertainties on the parameters. We may need to provide a generic wrapper that (a) ensures some uncertainty-aware object (from e.g. the `uncertainties` package) can be provided to the evaluation method in the model, and (b) return a `Spectrum1D` object with the defined uncertainties when calling the `fit_*` functions.

Start by reviewing the model evaluation methods and the fit_* functions described in the issue, along with the related issue #474. The work is complete when models with uncertain parameters propagate those uncertainties into evaluated output data and fit_* calls return a Spectrum1D object with defined uncertainties.

Written by the indexing model from the issue text.