Stop manually creating the list of tests (and searching missing). Main level test_all does a search for all test_* also possible to run by command line with starts
easier to spot missed tests (when guard and cmd-line differ)
seems to fit the layer much better as we really have a very reduced instruction set
