replaced add_definitions calls within PCLConfig.cmake.in with appending PCL_DEFINITIONS

[Algomorph]

PCLConfig previously used two strategies for providing compile definitions to the CMake targets dependent on pcl. Mostly, the defines were added to the PCL_DEFINITIONS variable, which then could be used in the dependency's cmake. In two instances, add_definitions was used directly.

The latter strategy causes errors in compilation of projects having some targets that use PCL and some that don't, while, for instance, using CUDA / nvcc. In the nvcc case, the error caused by extra compile flags is "nvcc fatal : A single input file is required for a non-link phase when an outputfile is specified"

Discussion:
It is unclear to me why compile definitions are necessary at all in this case. The file pcl_config.h.in is configured at the time of CMake generation for pcl, and it probably should have all the flags set correctly at that time instead of relying on -DDISABLE_<?> flags and such. Files that rely on these definitions should include the pcl_config.h header. This way, users don't need to use PCL_DEFINTIONS explicitly at all.

[SergioRAgostinho]

@nizar-sallem Can you have a look at this?

[taketwo]

+1 for the changes in this pull request.
Also, completely agree with the point made in discussion section. We may add removal of PCL_DEFINITIONS for the 1.9.0 list.