Implement saturated likelihood test for projections

[davidwalter2]

Implement saturated likelihood test for projections

• This is done using poi models where an unconstrained parameter is assigned for each bin in the projection. The fit is repeated and the saturated likelihoods of both fits is compared to extract a chi-squared distributed value and its probability.
• To be able to use it along other poi models I've created the CompositePOIModel class to compose multiple POI models. I've also added an init_fit_parms function to re-use the fitter object with a modified model

Updates for for data covariance

• Store data covariance in fitresult if present
• Added option '--dataCovariance' for plot hist script to plot data as line with uncertainty band to illustrate the fact that data points are correlated
• Add support for plotting data covariance (also for mappings e.g. projected to 1D) using '--dataCovariance'