Take the experiment algorithms out of the`__init__` methods

[drbenvincent]

At the moment the experiment algorithm all takes place in the __init__ methods. This is a bit un pythonic.

Step 1 could be to extract into a new method which is still caused in the init.

Step 2 could be to not automatically call that method when the experiment object is instantiated. Instead the experiment classes could get a fit method that kicks this off.

If we go with step 2, this would open the door to a more Bayesian workflow where we can explore the prior predictive before fitting, and change custom priors or model formulae first. Would need to think about it more, but not sure of this could be done just by creating the models on their own before sky experiment class is created. And I think that because the data only enters the model via the experiment class.