posterior importance sampling, which uses metropolis draws for linear model and computes PKF likelihood

handle heteroskedastic for plotting smoothed shocks

add wrapper, with regime search engine, for PKF kalman update, that applies to both filter and smoother.

distinguish between TolFun (optimizer termination criterion) and TolGstep/TolGstepRel, used for tuning gradient step.
So, we can keep for example TolFun=1.e-5, while allowing coarser values for TolGstep. this may help when maximizing non smooth surfaces (e.g. PKF or very large models), where numerical noise may count. By default TolGstep=TolFun as in usual historical behavior.

Manually resolved conflicts from 4.7 JRC branch

add utillities for foreast, irf, plot_irf and squeeze SD into occbin, and their relevant options [this commit must be better split before making a PR on gitlab]

new algorithm for weighted SD with occbin, where weights for the regime effect are common to all variables and set based on target variable, e.g. shadow rate in the case of ZLB.

(cherry picked from commit 09555f4b3461cfd31bfc411ee222aadb0c606707)

New option to slice that runs optimiser steps instead of slice draws: with this one can run optimisers in parallel, starting from different random locations.

(cherry picked from commit 6d03f78e)

[skip ci]

By the way, add some explanatory words.

[skip ci]

[skip ci]

occbin enhancements and bug fixes

See merge request Dynare/dynare!2123

- 321 when simulation within occbin smoother fails
- 322 when occbin smoother does not converge.

bug fix and avoid missing output arguments and when smoother exits after non convergence of simulation.

make sure also ALL updated variables in period 1 are reconstructed with smoother_redux. test model changed accordingly.

eliminate buggy and useless check for existence of guess regime variable. this may create bugs in running PKF update step!

adjust smoother redux test, to consider that now ALL smoothed variables must be properly reconstructed also in period 1