Question: Should the EK0 by default use a diagonal diffusion?

The issue: The uncertainties returned by the EK0 can be a bit surprising, as it only returns sacalar covariances (s^2 * I). That is, each output dimension obtains the same amount of uncertainty, independent of its scale. This has previously lead to confusion #25.

Proposal: Changing the default diffusion model to a DynamicMVDiffusion allows the EK0 to scale each output dimension independently. This could lead to (somewhat) more interpretable output uncertainties.

Related PR: #65 applies the proposed change.