Solves a bug with METAD + TARGET + GRID_SPARSE
See also #467

This commit contains changes automatically generated by plumedbot on Travis-ci.

cc: @valsson

* OpenMP changes to allow setting the number of threads from main code
furthermore avoiding to call getenv more than once (it is faster)

* setOpenMP: gromacs patch enabled to set the number of openMP threads

* OpenMP: moved static variables inside  a struct/function couple

* OpenMP: removed api interface and gromacs patch

* astyle

defined then just use one thread

@giovannibussi right?

(cherry picked from commit ae6c8ee4)

This commit contains changes automatically generated by plumedbot on Travis-ci.

Only for
RMSD TYPE=OPTIMAL

It looks like there were many incorrect simd statements. In particular:
- In all cases where a double was reduced, I added a reduction clause
- In all cases where a Vector was reduced, I just removed the pragma

This error lead to many failures in regtests when compiling with
intel -O3. It looks we have to make sure a compiler with simd support
is tested on travis-ci

Here I fix something I think might be wrong. In particular, if
safe==false && alEqDis
it looks like dist was initialized to the correct value but then not used.

From the current code coverage it looks this path was never tried.
Anyway, I am pretty sure the old code does not make sense.

I went through cppcheck warnings (style) and fixes those I understood.
Mostly issues related to scope of variables.

This commit contains changes automatically generated by plumedbot on Travis-ci.

I replaced a few pointers with vectors or arrays

(should go with af0def1e)

closes #244

[makedoc]

closes #268