ARG is just a compulsory keyword and not an input keyword

that appear in the manual.  I think this makes the manual clearer

more specific in the documentation as to what precisely it is doing.

I had to split BiasWalls into two separate files for upper and lower walls

this about

some of the English and in some cases writing new documentation.  Also sorted
a number of issues that arise because of the awkward formatting that is
Doxygen requires

action in html.

Fixed it so that the output of html also works for keywords that don't take
any input (for instance SetupUnits)

should not call the Action::registerKeyword stuff.

Indeed, since we stay on multiple inheritance, Action the keywords
should be added by the daughter classes

Added keyword registering functions to Action, ActionWithValue, ActionWithArguments and ActionSetup.
Also added variables so that the keyword list can be passed to the Actions

(merged from gat2-reorder branch)

Added keyword registering to all the action classes listed below

(merged from gat2-reorder branch)

Added register keywords functions to all the classes below
(merged from gat2-reorder branch)

Fixed the registerKeywords routines in the committed subroutines

Made sure colvar is adding keywords for numerical derivatives

Oops... genericDebug should register keywords for action pilot.  It does now

Added keyword registering for moving restraint

Added keyword registering for NATURAL keyword

COMPONENTS should have been registered as a flag and not an optional oops

no instances of this class are created anywhere

(merged from gat2-reorder branch)

fixed buggy loop in Keywords remove

Sorted out html alignment for keywords stuff.

I tried to simplify the Kearsley routines so as to take advantage of
Vector and Vector4d classes. I just did a scan of the file trying to
replace arrays consistently. It passes the regtets, so I think there
is no problem, and it runs at the same speed.

modulo(v), equivalent to v.modulo() and
modulo2(v), equivalent to v.modulo2().

They allow for syntax like modulo(v1+v2);

for a mistake I used physical copies instead of links on some regtests

I also updated the regtest and added a further case to it

In particular, I think there is some problem in the "fast" algorithm.

It tests different choices regarding SIMPLE/OPTIMAL and
regarding weights

This is just to debug functions with change their argument list
on the fly

To have it working I also had to update simplemd.f90 so as
to pass engconf to plumed

dipole to test the readin

It allows to declare vectors (tensors) of arbitrary
size N (NxM). The difference wrt Matrix and std::vector
is that dimensions should be declared at compilation time.
This allows to save the storage of extra information
(useful for small 3 or 3x3 objects) and to optimize the code.

To improve the performance, I used a single n*m array
for Tensors (instead of a [n][m] array). However,
the external interface is untouched.

Finally, I defined some typedefs to simplify usage:
Vector (for 3d, so as to maintain compatibility with the rest of plumed)
Tensor (for 3x3d, so as to maintain compatibility with the rest of plumed)

From the tests that I performed, there is no visible overhead
when compared with the older (fixed 3 or 3x3) vector/tensor.

inline function used before definition

Both file "forces" and "FORCES" were written, causing a stupid
problem on the mac. Moved "FORCES" to "ff" and added it to the regtest.

Probably we should better handle these cases and give an error...

Weights (align/displace) where set to zero. Now they are set to one.

I cannot understand why, but apparently there was a micture of COLVAR
and forces in the forces.reference file... Fixed now.

I had a problem with regtest rt6, probably due to numerical errors
on numerical derivatives. I added the possibility of specifying the format
in DUMPDERIVATIVES so that we can print just a few digits in the regtests.
As a default, I leave the "accurate" choice %15.10f.