Irf restrictions and identification

Fixed bad representation of identification patterns for moment information matrix (and also fixed figure title)

1) Define a new type of parameter Random_prior_params, when prior/posterior mean mode do not work and the code searches randomly in the prior space a parameter combination that provides unique saddle path solution;
2) save point estimates in separate file, in order not lose work when subsequent analyses are done for prior/posterior mean/mode etc.

write_latex_definitions: remove lines that prevented this code from being included in a larger latex doc

Closes #549

Fixes to mode_compute=5. (1) force univariate steps at every iteration: ...

Fixes to mode_compute=5. (1) force univariate steps at every iteration: improve robustness at the expenses of a bit less efficiency; (2) fix gradient computation for special cases when the value of the parameters get stacked at the prior boundary.

Trap the case where only one parameter is estimated and only one non-con...

Harmonize plotting of IRFS

Similar problem to #446

Trap the case where only one parameter is estimated and only one non-constant row is available in Jacobian: (1) do not use vnorm; (2) do not use advanced option.
(cherry picked from commit d9856879)

In stoch_simul.m IRFs were only printed if they were not flat instead of checking whether the IRF is different from 0. Unit root responses were thus filtered out, e.g. C=0.2 for all t was not displayed.

Document treatment of parameters in steady state files

Changed documented behaviour of @dseries/save method (do not document the default value of basename).