Refactor package

[lmh91]

I am currently refactoring the package.

The main difference is that model functions now have their own struct similar to
distributions, e.g. Normal, from the Distributions.jl package.

Currently, there is one predefined model function, NormalPeakUvD, which is a Gaussian plus linear background model.
But the user can of course also define his own models, see Doc.

Also, the fit syntax has changed. It is now like the fit function from StatsBase.jl and Distributions.jl:
fit(<::Type{Dist}>, data)::<Dist> where in our case the data are StatsBase.Histograms:
e.g.: fit(NormalPeakUvD, hist) which would return an instance of NormalPeakUvD with the fitted parameters.

Two fit "backends" are implemented:

• Maximum LikeLiHood Estimation (MLE) via Optim.jl
• Bayesian Fit via BAT.jl (optional via Require.jl)

They can be chosen via a keyword: backend = :optim (default) or backend = :BAT.

Both methods use the gradient of the model functions via AutoDiff.
User specific model functions need to be defined such that ForwardDiff.jl can differentiate them.
Currently, there is not yet a fallback to a non-gradient based optimizer / sampling algorithm.
But this can be easily added.

Also, NamedTuples can be used for the parameters of a model.
And the parameters can be individual set constant via ValueShapes.jl
such that they are fixed in the fits but the model does not have to be redefined.

[lmh91]

A preview of the docs can be found under
https://juliaphysics.github.io/RadiationSpectra.jl/previews/PR36/