We would like to implement a version of the DLM prior where
beta_t = alpha_t + u_t
alpha_t = alpha_{t-1} + delta_{t-1} + v_t
delta_t = delta_{t-1}
This should be relatively simple, since it's a linear model. When we do this we need to prevent users from setting phi != 1. We should also come up with a consistent interface for this model and the model
beta_t = alpha_t + u_t
alpha_t = alpha_{t-1} + delta_{t-1}
delta_t = delta_{t-1} + w_t
which is currently specified using 'level = NULL'.