this is done supporting the combination of type == 2 with style == 0,
i.e. a local scalar, which is not available in C++ (but there one can
access the compute style data member directly. for the python interface,
the pointer is automatically dereferenced and returned as a c_int.

PyLammps now supports thermo_style entries such as c_msd[4]

The library interface was extended to provide direct access to atom data using
numpy arrays. No data copies are made and numpy operations directly manipulate
memory of the native code.

To keep this numpy dependency optional, all functions are wrapped into the
lammps.numpy sub-object which is only loaded when accessed.

Also fixes some regressions due to command syntax changes

The interface of that function has changed and includes two additional
parameters, which haven't been added to the Python interface either.
This showed up by trying to run the simple.py example.

Based on material presented during MD Workshop at Temple University in
August 2016.

Thanks to Giacomo Fiorin for figuring this out with NAMD/Colvars.
This requires linking with -Xlinker -export-dynamic or equivalent,
which is the default when using python-config to provide linker flags.
We will fall back to loading the DSO in case the initial load fails.

0aebb2eabea6ae3f5365951335ac3d698ab9ad9e

git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@15668 f3b2605a-c512-4ea7-a41b-209d697bcdaa

git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@15634 f3b2605a-c512-4ea7-a41b-209d697bcdaa