Incidentally, this means that the JSONlab submodule is no longer needed, so it
is removed in the present commit.

Closes: #1907

Replace it by nansum from the statistics package.

MATLAB R2018b does not have a true batch mode, and will not exit if there is no
final quit statement in a script. Hence ensure there is one when testing a .m
file.

By the way, remove a batch argument for Windows that is obviously incorrect.

On that version, spaces in variables names within a call to “table” are not
supported.

Ref. #1907

Ref. !2210

As a consequence, drop old uncrustify config file.

Indexes all C++, Fortran and C files (including submodules).

The output is always useful (and even needed in case of test failure).

manual: port changes from Dynare/preprocessor@e85d085a

See merge request Dynare/dynare!2208

NB: does not (yet) works with Occbin regime-specific equations.

Ref. #1890

Ref. preprocessor#104

model_diagnostics.m: add check for non-zero exogenous with OccBin

See merge request Dynare/dynare!2206

Information is rarely helpful but clutters output of model_diagnostics

Kalman filter mex: remove compilation warnings

See merge request Dynare/dynare!2207

Compilation wargnings were pointing at: (i) some supposedly uninitialized variables (-Wmaybe-uninitialized); (ii) a pointer assignment (=) to a non-pointer target (-Wsurprising).

This replaces the old hack which was using a local Linux-specific symlink. Now,
the Dynare default is the same as in Sphinx, i.e. rely on a CDN for delivering
MathJax.

Adapt our CI accordingly to continue using MathJax from the Debian package
there (to avoid privacy issues related to the use of a CDN).

 - Ensure that we always use the same DGP (e.g. same transition matrix).
 - Call the mex more than once with different samples.
 - Ensure that the size of the state space model is the same in tests 1, 2 and 3.
 - Fix the seed (same samples across 1, 2 and 3 except for the additive noise in 2 and 3 on the observed variables).

The latter is now much faster than the implementation in the bytecode
MEX (which this commit thus removes).

The maximum error is computed over the change in endogenous (i.e. ΔY).