Add options for providing the variances, an identity matrix , and user specified matrices as proposal densities

Deals with #507 and #112
Includes a unit test

Changed the organization of the options for the dynare's implementation of the simplex optimization algorithm.

Changed the definition of the maximum number of function evaluations in dynare'es iplementation of simplex algorithm (mode_compute=8).

Changed the organization of the options for gmhmaxlik (mode_compute=6) so that options can be set using the optim option of the estimation command. Added an option (targeted acceptance rate).

Implements #423. To be done: add preprocessor options for tolf and maxit.

Added an option for setting the number of points where the posterior kernel is evaluated when using mode_check. Reorganized the options_ global structure wrt mode_check.

mode_check_symmetric_plots (=1 at default, i.e. current behaviour)

Implements and documents the Andreasen et al. 2013 pruning at order=3
and sets it as the default at this order. Michel's pruning based on the
approximation of the forecast function has been assigned the option
pruning_forecast_approximation. The preprocessor-interface still needs
to be added for this option. Moreover, more documentation/a reference
what this option does is needed. At a later point, we might change the
default to Michel's approach.

1. The first call to set_prior overwrote the first column of
estim_params_.var_endo storing the position of the variable with
measurement error in M_.endo_names with the position in
options_.var_obs. All subsequent calls to set_prior then lead to
crashes.
2. At the same time, for correlations of ME, the first column of
estim_params_.corrn still stored the position of the variable with
measurement error in M_.endo_names. But subsequent calls to it were done
as if it stored the position in options_.var_obs

I introduced two new variables in estim_params_ storing the respective
positions in var_obs so as to not necessitate changes in the
preprocessors.

3. For cases of calibrated measurement error correlations, the
covariance matrix was not updated.

4. Fixing a lot of smaller bugs related to measurement errors, including
some copy and paste errors
-

Add option and code for endogenous priors according to
Christiano/Trabandt/Walentin 2011, JEDC. Still needs to be integrated to
manual and pre-processor.

This reverts commit cacb0470.

It was making bytecode crashing. This field will still have to be removed later
when dealing with #295.

-add a variable to make a difference between gaussian approximations during importance sampling and approximation of states (for gaussian and gaussian-mixture filters)
-modify the default of parameters for the gaussian approximation using the scaled unscented transformation

- fix support for filenames with extension (.m, .mat, .xls), since the manual
  and the bison grammar both support it
- fix support for XLS files under Octave
- add support for XLSX files
- run test for loading XLS files

ms-sbvar: remove unused field removal statements (and set options_.datafile='' in global_initialization)

Added an option to decide if dsge_likelihood should call univariate filters when the covariance matrix of
the prediction error is singular (default is yes).

Added the possibility, for each given size of the sample, to restart an arbitrary number of times the estimation when estimating a model with a recursive approach.

Deactivated by default (options_.recursive_estimation_restart is defined to be zero in global_initialization.m).

(cherry picked from commit 4053f664ecec14ffb7269c6bbf78355ca46b6b92)

options_.gpu (no interface yet, use options_.gpu=1). Added test
./tests/first_order/fs2000_cr.mod

dyn_first_order_solver.m in global_initialization.m This is necessary
if noclearall option is used.