EM algorithm for linear state-space models

Computes the MLE parameter estimate for linear state-space models with the EM algorithm.

Computes the maximum-likelihood estimate of A,B,R,E,F and Q in
Y(:,t) = A + B*X(:,t) + e(:,t), e(:,t)~N(0,R)
X(:,t) = E + F*X(:,t-1) + u(:,t), u(:,t)~N(0,Q)
where Y is a N by T vector of observables and X is a K by T unobserved state vector.
A structural parameter-expanded EM algorithm is used for computing one element of the parameter set estimate which is mapped to the unique point estimate in a normalized parameter space. Many popular normalizations (parameterizations) are supported. The algorithm implements a square-root Kalman filter. Overall, the SPX-EM algorithm is more robust and converges much faster than a standard EM algorithm. This first release has few bells and whistles: Please let me know of any additional features that might be useful to you. Report bugs or unexpected behavior as well...