TODO list
 * Add argument check.result for function ldl which warns if the decomposition was not successful 
   (due to bad tolerance parameter).
 * Add more tests, at least for structural time series, ARIMA and mixed models and multiple tests 
   for SSModel and auxiliary functions.
 * Study and possily implement marginal likelihood correction used in Francke et al. (2010)
 * Check and correct documentations.

TODO in distant future:
 ** Move everything from Fortran to C++ with Rcpp and start using S4 and sparse matrices...