relative paths

Setting the PYTHONPATH in ost_config is no longer required (This was
breaking stuff for projects relying on OpenStructure).

It's causing problem on some systems

This allows to set the title to an empty string

The scene background can now be a gradient or a bitmap image in addition to a static color; use scene.SetBackground(gfx.Gradient) or scene.SetBackground(gfx.Bitmap) for this purpose.

The gfx.Bitmap class and its import/export functions have been exported to Python.

Some cleanup in the OpenGL code to (i) avoid premature rendering before full initialization and (ii) streamline error checking a bit.

the internal representation of a transformation for EntityImpl was changed from Mat4 to Transform
in preparation for adding transformations to entity views; this impacts quite a few higher-level routines
that used the transformation matrix, which now can use the more elegant Transform::Apply method instead;

GetTransformationMatrix and IsTransformationIdentity methods in the EntityHandle interface are marked
as deprecated; use GetTransform and HasTransform instead. In addition, there is a ClearTransform method
which ensures that the internal has_transform flag gets set to false

geom::Transform was extended to offer ApplyInverse, and matrix inversion exceptions are handled internally

constructing a Vec3 from a Vec4 no longer throws a DivideByZero error when w is close to zero, instead it
is silently assumed that this is not a true homogeneous coordinate conversion

as an additional comment (and note to self), atom->GetPos always returns transformed positions; however,
GetAltPos returns original positions, and original positions are also stored in the AtomGroup on the
Impl level. Both the mol unit test as well as the PDB writer expect this behavior; the PDB writer applies
the transform after calling GetAltPos. Commented code is in place now to change this behavior, i.e. have
GetAltPos return transformed positions, making it consistent with GetPos, should this be desired.

fixed Transform SetMatrix behaviour
fixed Transform unit tests
fixed mol.Transform proxy
added Transform.Apply(Transform)

behaviour in GaussianSmooth function in table class

Instead of doing a.GetFloatProp('foo'), feel free to use
a.foo. a.foo = 3.0 will create a new float property 'foo'.

This commit is brought to you by ZSO Pilatus

to the editor. still messy and no unit tests...

corrected in-code occurences of mol/transform.hh, mol::Transform, and mol.Transform to geom equivalents

The mol/transform.hh header still exists and typedefs geom.Transform to mol.Transform; however, including it
will give you a deprecation warning; please observe these warning in your code and change your Transforms to
geom::

The Transform class was moved from the mol module to the geom module. A typedef
was put into the mol namespace in C++, and a similar alias was put into the
mol python module, with an added deprecation warning.

The idea was to speed up the build process by removing several
unneccessary targets and stage all headers in one go. However,
for some reason, CMake does not think that's a good idea. Only
parts of the headers get staged