The idea is simple: each print represents a testcase within a test unit.
As we don't have strict rules on which/how many testcase are put into
a test file, it's nice to have an idea how many *testcases* we have
totally. Would be nice to count how many testcases pass/fail, but
that's a bit less trivial.

tests/bytecode/run-tests already uses puthon3.3, so let's just use it here
too.

Fore reference, errors with python 3.2.3:

  File "tests/generator1.py", line 12
    return None
SyntaxError: 'return' with argument inside generator

  File "tests/list_clear.py", line 3, in <module>
    x.clear()
AttributeError: 'list' object has no attribute 'clear'

etc.

import works for simple cases.  Still work to do on finding the right
script, and setting globals/locals correctly when running an imported
function.