(cherry picked from commit ae6c8ee4)

(should go with af0def1e)

closes #244

Since matheval evaluator is not thread safe, it is necessary to allocate
one evaluator per threads

Closes #249

Notice that almost all files were updated with the astyle commit
2813d759

Year in header is here updated to 2017

This is a huge commit, applying indentation to the whole
code. Notice that this commit is made on master branch
after having merged with -s ours from branch v2.3

See #215

This is a huge commit, applying indentation to the whole
code.

See #215

Solves an issue when using switching functions without specifying DMAX
leading to a crash with some particular compiler

When making mistakes such as
STRIDE=2.5
or
COORDINATION NN=6.5
It gives an error instead of using the default values.

This fixes #104

Using common routines for input of atoms across all MultiColvars, MultiColvarFunctions and AdjacencyMatrices

related to #137

It is now possible to specify a generic switching function
with MATHEVAL. Notice that this is slower than the C++
implementation. Anyway, it allows arbitrary switching functions
to be quickly tested.

MM=0 now implies MM=2*NN. This means one can make a steep function with
COORDINATION NN=1000
without taking care of MM.

Fixes #137

this is affecting a lot of regtests because of small numerical
changes.

Since in the simple syntax (COORDINATION NN=xx) there is
no way to control stretching, I cannot change the input files
to keep output unchanged, so I had to reset a lot
of tests

I am not sure whether we should merge this

I did it only when using the advanced syntax (SWITCH={...}).
I changed the input in some regtests to comply with
this change.