* Marco De La Pierre: added WALKERS_DIR option for WALKERS_MPI

* Marco De La Pierre: WALKERS_DIR option for WALKERS_MPI, not for single replica runs

* Marco De La Pierre: WALKERS_DIR for multiple walkers metad made general

[makedoc]

Dimer CV

* Citation fixed as suggested
* Added PBC-friendly virial contributions: compared with the numerical derivatives
  for a configuration and the same with a dimer shifted by the lattice vector on one of its components.

This is required since intel compiler does not have complete c++11
support but still works correctly.

Perhaps we have to find a way to include intel compiler on travis
so as to be sure we do not include unsupported c++11 features.

[makedoc]

the regtext might produce different results on other architectures so I added
a numeric format keyword to the plumed.dat file of both regtexts.
Also the test .pdb configurations are improved, being now taken from a trajectory rather than
a starting configuration.

removed mention to a bug that is not actually there

Added other failing examples for the regtest based on exceptions.

[makedoc]

In case of error with LOAD, the code was returning an error without
raising an exception.

This should make the code slightly fast.
Notice that it is now necessary to explicitly sort the watches
when printing the log.

These files were related to the original implementation of RMSD.
I think they have not been used for years, so I remove them.

[makedoc]

This is done only in the cmd interpreter

This should make the code slightly fast.

[makedoc]

Now all installers use a temporary directory