in cppcheck job, actually run make codecheck

In the first job on travis-ci I set PLUMED_ALL_TESTS=yes

With this setting, if some of the regtests is not run the whole
test suite fails. This actually checks that all the optional modules
have been enabled at least in the first job of travis-ci

It works creating a local Portfile so as to test
non released plumed versions.

Pdf manual takes a lot of time to be compiled.
Since it is not very useful, I disable it by default.

Notice that on travis CI it is enables during configure.

[makedoc]

[makedoc]

The new message class (information) is treated as (style).

I temporarily add two travis test to allow 1.71 to fail.

[makedoc]

I modified extensively the travis file to allow multi-os support
and to use more recent UBUNTU:
- I upgraded most of the tests to UBUNTU trusty (with gcc 4.8)
- For a couple of tests I switch to UBUNTU precise (with gcc 4.6)
  to be sure we keep compatibility with older compilers
- I removed clang tests from UBUNTU and added OSX tests (both serial and MPI)

Since UBUNTU trusty comes with doxygen 1.8.6, I install doxygen 1.8.7
from source. Notice that this was solved in UBUNTU precise by using
an external ppa database. I prefer install from source here
since in this way we can easily upgrade doxygen to use latest features.

[makedoc]

Some time ago I made one build allowed to fail.
After c9e9c366
(see also #183) it looks like the issue is solved,
so I put the build back in place.

more tests

[makedoc]

[almost]

[almost]

[almost]

It seems in 865d16c9 I introduced a bug:
checks where reported to travis only when
building doc.

Fixed now

I changed from , to : separator which is more standard
(e.g. in PATH)

Added an option to ./configure to choose modules.

./configure --enable-modules=all
enables all optional modules

./configure --enable-modules=none
disables all optional modules

./configure --enable-modules=reset
reset default modules

./configure --enable-modules=+manyrestraints,-colvar
enables manyrestraints and disables colvar

./configure --enable-modules
maps to ./configure --enable-modules=reset

./configure --disable-modules
maps to ./configure --enable-modules=none

Now $HOME/opt has the priority. This allows properly
overriding already installed programs.

In 84ad8d2a a bug was introduced and
manual was not correctly pushed from travis-ci

Now in the cppcheck plumed is not built or tested.
Only cppcheck is performed

This reverts commit e641cbb7.

(now I recall why I setted openmpi1.5: openmpi1.3 makes a regtest
fail randomly)

I think the only failing case was openmp+mpi+debug options.
I disabled openmp in that case, so that it should work. Let's see...

(related to c933ced1)

For some reason one of the setting of travis is failing
on rt-dfg2. I don't understand why, but I suspect it comes
from the internal lapack. I try to change the test to use external
lapack and I keep as a record the test that was failing
as "allowed_failure". Perhaps someone else will find the time to
understand where is the problem.

Notice that on machines where internal lapack were used
./configure should be run again.