refactoring

- rewrite kenStone and duplex in C++ and avoid computing the whole distance matrix at once (Several users complained that they would run into memory problems for bigger datasets (> 20-30 k obs)). This is should be fairly easy to do
