Journal extension and additional improvements

Big update in the repository, including:

• 3D computation of the distance maps, done at the pre-processing time;
• other losses implementated, for comparison;
• example of a stand-alone, multi-class setting (ACDC dataset);
• improved readme and instructions;
• updated for latest python and pytorch releases;
• removed remnants from our constrained-cnn works (previously a shared codebase);
• add colors to the makefile, for improved readability;
• more flexible makefiles, that allow separate results folders (through the RD environment variable);
• submodule for the viewer, pointing to https://github.com/HKervadec/segmentation_viewer ;
• the training recipe now store commit-hash and diff, to track exactly the code used;
• (partial) list of papers using our loss in their method.