Corrected bugs in the BNN and added many plotting scripts adapted for the paper

• Corrected bugs in the BNN and moved relevant code from torchbnn to remove its dependency.
• Split log_sigma per layer, so that one can do proper empirical Bayes.
• Save CSV files in offline_analysis.py, so that we can plot them offline with prepare_plots.py.
• Added several details in the PES plot, to provide more details on the PES measurement structure.
• Analyse results of the model fitting as a function of number of channels to estimate the impact of the number of channels in the resulting resolution.
• Fixed bug when number of channels to read do not include all.
• Allowed setting of number of peaks.