Hand-made list of tests is not great. As Christopher suggested we can find
all executables in a directory and treat them as tests. This idea is good,
but requires a little bit more trickery.

First, some executables in directory are not tests, e.g. these are per-test
scripts, which should be skipped. Next, not all tests in zdtm can be right
now run in continuous manner (they fail), but we want to run them manually.
To fix that such tests are marked with 'noauto' flag in the description.

So we have the test test descriptor file, which states in which file to
look for tests (executable) and which of them to exclude. Maybe more, can
be added on demand.

And, finally, the per-test description goes in a ${test}.desc file in native
python eval/repr-able format. If the file is absent a default description
is used.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>