PlumedFile's were not flushed correctly

Attempt to solve the problem of incomplete HILLS file. Seems to be working, but need some revisions and discussions with you guys…

I added a method to reset the EOF indicator of a stream. Needed by multiple walkers. Still is not working properly when one walker is reading an incomplete line (because it is being written) from another walker's file.  In these cases, I got:
+++ Internal PLUMED error
+++ file PlumedFile.cpp, line 311, function PLMD::PlumedIFile &PLMD::PlumedIFile::advanceField()
+++ message: assertion failed nf==words.size()

Initial commit for multiple walkers. Still not working. There is something I don't understand in the PlumedFile stuff. Why can't I create a std::vector of PlumedIfiles?

It was still consistent with the older implementation (the one requiring
addField()).

So that it can be used for testing

While trying to use the new printField stuff for metadynamics I
realized that it is simpler if one does not need to explicitly
declare which fields will be printed.

I also fixed the formats in two regtests

I added a class for reading (PlumedIFile),
fixed a few problems in the PlumedOFile, improved
documentation. Still it is not completely tested.

File is "linked" to the action, so that it imports properties
such as the communicator etc.

Documentation is still a bit poor. I used it to replace
the Log implementation and also for in the GenericPrint class.
In the log, this implies a small change (FIELDS line is
written in also in the log).

I tried to make everything so that regtests are not affected. Only,
change, now the rt17 (using _ROTATE) works correctly, writing
field names as they are updated.

Curly brackets {} should be used instead of parenthesis () to delimit input
parameters. This will allow to use parenthesis in matheval.

Regtests and documentation have been updated accordingly - I hope there not
inconsistencies around!

I try to use forward declaration as much as possible to decrease
compilation dependencies

The proper way to disambiguate with STL is to use std:: namespace

As a general rule, when one adds a file he should run
./header.sh
This will add the LGPL header to all the sources. 

Let's try to keep consistent names

The two indexes can be of different type.

to allow the bias-exchange like patterns

of setting the seed from the plumed.dat files. iy=0 in Random.cpp

Fixed derivatives for Ghost Atoms. At least I hope so. Is there a way to to numerical derivatives of ghost atoms?
If not, we need it!!! Max