Some changes for the PyTorch IBA API

[berleon]

A few remarks to the PyTorch IBA API:

• IBA.heatmap rename to IBA.analyze. Add flag to switch between saliency_map (in bits, channels summed, scaled to input image) or capacity (in bits, not summed, not scaled)
• IBA.supress_information rename to IBA.restrict_flow
• IBA.__init__ add estimator=None parameter
• IBA.__init__ add feature_mean_std=None parameter which takes a tuple(mean, std)
• IBA._init rename to IBA._build: make clear that it is different to __init__