Expand the set of example runcards to showcase more diverse applications

[jmarshrossney]

E.g.

• Checking bare mass is reproduced when training against non-interacting theory
• Studying intermediate layers as probability distributions
• Looking at the weights and biases of neural networks
• Measuring the integrated autocorrelation time by forcing sample_interval=1, versus allowing it to be set automatically to speed up generation of a representative sample
• Showcase different flow models
• Compare batch normalisation with global rescaling (fixed / learnable scale parameter)