Kalman filter log-likelihood
[llik]=filter_like(param,func,y,x,F,z)
* param = a vector of parameters (sqrt of variances)
* func = the function which transforms the parameters into the matrix of variances (Q and R)
* y = (nx1) data vector
* x = (nxk) data matrix
* F = transition matrix
* z = (nxl) data matrix
* llik = - 2*log-likehood (+n*log(2*%pi))