What should be the default EpiAware metric shape be?

[SamuelBrand1]

At the moment our default is that NUTS sampling has a diagonal metric shape.

Should we change this to DenseEuclideanMetric by default, given that this approach might be pretty inefficient if we posterior correlation. This is associated with #238 because adding pathfinder metric makes it easier to estimate a "good" dense kinetic energy metric.