* Added stretch widget to PropertyCollectionWidget

CudaConfidenceMapsDemo pipeline now has the feature to collect
statistics about the computed frames. The statistics can be exported in
CSV format to the clipboard.

Label is now hidden by default for Button Properties which saves space
for long button descriptions

added support for fixed iteration count to ConfidenceMapsSolver again.
It is now possible to chose wether to use a time budget or a fixed
iteration count.

You can now use the F keys (F1 - F5) to change presets in
ConfidenceMapsDemo.

* ConfidenceMapsDemo pipeline can now interpret voxel size as fan
  geometry. This is actually a quick quirk. The fan geometry is encoded in
  (x, y, z) as (fan-angle, inner-radius, 1.0f)
* Added support for the current cusp version on git

* Remove explicit use of Qt::QueuedConnection. As it turns out, when the
  event is emitted by a thread, different from the GUI thread, it is
  automatically enqueued.
* use #ifdef instead of #if

* OpenIGTLink processor now is robust against missing voxel size in image
  packet.

* CMake has been improved and now supports for example module dependencies
  and works for visual studio compilations
* An image cropping stage has been added to the pipeline to trim any
  garbage that might show up at the border during image acquisition.
* Added support for 2D images to the GlImageCrop Processor

the solver no longer stops early if the convergence criterion is met,
instead it always executes the specified number of iterations.
This allows the code to skip the residual norm computation and copying it
over to the host on each iteration and results in significant speed
increase.

CudaConfidenceMaps now support a 4 neighbourhood version. This results
in a smaller number of weights and should theoretically be faster.

* Removed some dead code.
* Avoided one texture copy in the ConfidenceMapsSolver Processor
* Exposed one more option to the ConfidenceMapsDemo piepline

this avoids having to const_cast constant buffers before passing them as
input o the function

* Now it is possible to set a time budget for the CudaConfidenceMapsDemo
  pipeline and the number of CG iterations that can be carried out is
  automatically determined.
* Code still needs some cleanup.

* this is achieved by avoiding allocating a chunk of memory for the image on the
  GPU each time, but rather reusing the same buffer over and over.
* Added some comments to the solver code

attenuations were not computed correctly

On my machine this means a performance increase from 8ms to <1ms. Time that can
be spent running more iterations of the CG algorithm to get more accurate solutions.

Switch wrapping mode for texture in fan renderer, so that colors don't
bleed to the other sides on the edges.

Now the pipeline correctly uses the fused confidence image as input
for the fan geometry rendering.

The texture has the same size as the original us-image