random tester

1) we need a data set
2) then a program doing random tests with it (choosing a command randomly then
choosing more or less randomly the command parameters)

If it can run long enough, it will probably do a full code coverage or discover
bugs if stopping prematurely
