When the log file is chosen by name (cmd("setLogFile"...))
there was a problem due to an apparent double initialization
of log file. Now, logfile is open when cmd("setLogFile"...)
and, in case it is still not open when cmd("init") is called,
it is opened to stdout

For real numbers, one should use fabs

I have greatly simplified the hierarchy of classes that inherit from Vessel.
Furthermore, I have re-used much of the parsing functionality in Action so as
to make adding a Vessel more like adding an Action. All the changes in regtests are
just because of changes of the names in the header line.

In addition I added functionality so that multicolvar can pretend to be colvars
Multicolvars have value that are always labelled as label.something.
However, in AlphaRMSD etc the default is to calculate the number of
RMSDs that is less than a certain target value.  Furthermore, in the
manual these MultiColvars are listed as colvars.  The multicolvar functionality
inside alpharmsd being a feature for "advanced users".  Hence, to avoid confusion
I added something so that by default the value in alpharmsd etc is just
label.

This allows using less digits in regtests. I also updated all the regtests with METAD, including some *mpi tests which were failing because were not
updated after a recent change (removal of hill at time=0)

This implements sum_hills tool via command line
it is called via a cltool that creates a
fake set of CVs and then creates an function actionwitharguments
which acts as hills integrator in the initial phase
(no calculate phase. It dies immediately). Hypothetically
this structure should allow also an integration  of the hills
while doing the simulations (not yet implemented).
It makes use of the Kernels instead of the internals of metad.
(TODO: replace meta with kernels)
See the documentation on cltool/SumHills.cpp

So it implementes:
added FakeCv to be used with sum_hills
sum hills accepts  stride (--stride) projections (--idw) via names
automatically grid sizing (beware, sometimes is crazy),
histogram correction for multivariate gaussians (--histo)
A one-line gaussian reader for metadynamics (more flexible)
A BiasRepresentation class which is intended to keep the list of kernels and
the grid representation in a single bunch. Can be useful when replacing
metad with kernels.
TODO: when periodic automatically sets the boundary. this must be fixed in a more
clever way
TODO  when projecting handle the derivatives so to have a grid that is usable for restaring on a
lower dimension

with plumed1

Hill in metadynamics is not dumped on the first step now. More consistent with plumed1 behaviour.

This I found to be helful when debugging on the cluster with buffered output.

This is an attempt to make it easier to re-use ActionWithVessel for things other than
MultiColvar. There are multiple ways to do a neighbor list and so I felt this functionality
should be separated somewhat.  In addition the way I had implemented the neighbor list for
multicolvar meant that the list was being updated one step later than it should have been.
Also did some tidying of the multicolvar and vessel code

- the patch.sh and src/lib/Makefile are the same of those in the branch tmp-cmake

the CMAKE patch works different from the former gromacs patches:

- untar gromacs-4.6b3 (./gromacs-4.6/)
- in gromacs-4.6/ copy patch.sh and patch
- mkdir a build folder ./gromacs-4.6/gmx-plumed/
- in gmx-plumed/ run cmake ../ to configure gromacs
- make / make install as usual

there is a bug somewhere for which a "-l" is added before
Accelerate in "-framework Accelerate" thus resulting in a
compilation error.

This means we avoid having to transfer data from the buffers in the
individual vessel buffers to the ActionWithVessel buffer for mpi gather.
The code might be slightly faster as a result

The multicolvar way of doing DHENERGY is particularly useful if you want
to calculate multiple instances of the DHENERGY with different parameters.
By using the multicolvar you can avoid the recalculation of the distances
for each different parameter set. You can also use the multicolvar to calculate
COORDINATION (LESS_THAN) and the DHENERGY for two groups of atoms at the same time

In addition I made some small changes to Vessel base class.  When numbered keywords
(WITHIN1, WITHIN2... )  are read in the number (the 1, 2... ) is also passed to the
vessel.  This makes it easier to create different value names for the various values
calculated by different instances of vessels of the same type.

If you want to but a plumed_assert in a loop you should now use
plumed_dbg_assert / plumed_dbg_massert.  This will not be included
in the code if you compile with -DNDEBUG

I think we should always use assertions (as now they also detect most of
the user errors). Anyway, in this manner it is possible to undefine them
to monitor their impact on performance, which is expected to be negligible.