This allow to have only one application to handle bot supervised and
unsupervised classifier.

Classifier category is retrieve with selected classifier and depending
this category some operation are added/removed for the training.

This workaround is added to avoid a crash in application when build
is made with RelWithDebInfo which is caused by a
boost::archive_exception impossible to be catch.
This workaround break the compatibility with shark model direct
read/write.
It is still possible to read old RF model from OTB.

SampleSelection can now be used without input vector data.

When only image is provided to the application a default vector
corresponding to the image envelope is generated.

All Supervised and Unsupervised Classifier should be declared and stay
in Supervised/MachinLearningModelFactory since multiple application
depend on it.

Use inheritance with otbTrainImagesBase to create unsupervised
composite application.

Create a common class with template to perform Supervised classification
(same as otbTrainImagesClassifier) or Unsupervised.

Add the ability to specify Supervised or Unsupervised classification

Inherit TrainVectorClassifier from TrainVectorBase and use Non-Virtual
Function Idiom to provide common behavior for Unsupervised and
Supervised classification.

Move commons files for both supervised and unsupervised into
LearningBase Module.
Also update Test and includes.