Return the whole simulated paths instead of only the contemporaneous reaction of the endogenous variables.

Otherwise we can not test the number of periods and we cannot use the previous solution
to initialize the following (stochastic) perfect foresight problem.

When a deep parameter was initialized to NaN (this is the default) any update of this parameter
done in the steady state model block was not propagated to M_.params. The problem was caused by
the definition of updated_params_flag.

Added an option fot the mode_check plots for defining the size of the neighbourhood around the estimated posterior mode.

Bug fix. When the estimation command is used, do not issue an error message if options_.order>1. In this
case reset the options_.order equal to one (except if options_.particle.status==1).

to set the seed of the random number generators).

Fixed typo. Do not use the non linear filter automatically when order==2, otherwise the non linear filters are used if the estimation command is used after a stoch_simul command with order==2. The non linear filter is used only if options_.particle.status is explicitely set to 1 (before the estimation command). A new estimation option is needed...

and order ~= 0 solve_stochastic_perfect_foresight_model().
checking only first period of simulation for convergence when checking horizon

Moved CMAES options in global_initialization, so that the options of this routine can be modified by
the user by writting in options_.cmaes.

Added missing field in bayestopt_ (restrict_var_list, indices for the state and observed variables).

Updated test on options_.order (current version can estimate dsge models with non linear filters considering a second order approximation of the model).

equations are approximated at order two around the deterministic steady state.