runs Kalman-Bucy filter over observations matrix Z
for 1-step prediction onto matrix X (X can = Z)
with model order p
V = initial covariance of observation sequence noise
returns model parameter estimation sequence A,
sequence of predicted outcomes y_pred
and error matrix Ey (reshaped) for y and Ea for a
along with inovation prob P = P(y_t | D_t-1) = evidence
