[Do not merge] Extend post-processing and plotting to act on intermediate layers of model

[jmarshrossney]

Phase 1:

• Action that yields model parameters (weights and biases) layer-by-layer
• Action that yields batches of configurations layer-by-layer (including base and Metropolis output)
• Implement basic plotting
• Subplots to separate weights from different networks, if interesting

Phase 2:

• Add runcard parameter layer_id to allow looping the entire analysis (including correlation functions, bootstrapping etc) over each layer in the model
• Tables + save to .csv as well as the usual plots
• Write script that reads in data from saved .csv and plots all of the layer on a single figure

Ambitions:

• Instead of just numbers, it would be nice if layer_id could refer to an attribute of the layers in the model. E.g., if I wanted to plot the spline layer and the MCMC output I could write layer_ids: [spline, metropolis].
• Better labelling of layers in the report.

[wilsonmr]

Just the final commit that is

[jmarshrossney]

Ok done. Didn't realise that was the breaking you were referring to.

[wilsonmr]

My fault for not being specific enough! If we get that merged soon then we can rebase this

[jmarshrossney]

I will need to adjust this after #68 is merged, since the model is no longer necessarily a flat Sequential of coupling blocks.