Configuration and output clean up

[kazewong]

1. Clean up configuration dictionary
2. Fix output format to interface with arviz

[kazewong]

For configuration, we should look into different options for configuration.

I think .yaml and Hydra is pretty good for this purpose https://hydra.cc/
Once we have decided, we can move forward in moving the config dictionary out of the example script.

[dfm]

I have so many opinions about this. I am confident that yaml and hydra are not the right choices... Happy to chat more!

[marylou-gabrie]

Concerning the output - we should also think of the right data structure to collect all of our outputs. Desired outputs already mentioned:

• chains (-> which format? One of the traditional sampling package, including to have named variables?)
• NF learned weights
• global/local acceptance
• NF loss over training iterations

[kazewong]

I have so many opinions about this. I am confident that yaml and hydra are not the right choices... Happy to chat more!

@dfm I don't have strong opinion on how to do this, happy to learn about different options!

[kazewong]

In branch samplerCleanUp commit a473131,
the configuration parameters are moved within the sampler class.

It doesn't necessarily shorten the examples, but the variables are more well defined since they are variables within a class instead of key-value pairs in a dict.

Tested in dual moon example.

[kazewong]

Sampler clean up seems to be done with #10. Closing this issue.