Fixed in r3347.

In r3348, I also added some more graceful handling of the case that a result dir
for ./experiment.py already exists. In that case, we now don't attempt the
problem. Instead, we just skip it and print an error message. Error messages are
repeated at end of experiment output so that they are harder to miss.

Before r3348, experiments whose results already existed would be attempted, but
there would be an error after the planner run which would cause the experiment
to be aborted and require manual cleanup.

Running an experiment using experiment.py produces the following error:

[2009-10-14 17:29:12] #0: Solving blocks:probBLOCKS-4-0.pddl with CausalGraph...
[2009-10-14 17:29:12] There were critical errors. See details on stderr.
[2009-10-14 17:29:12] critical error: temp dir
tmpdir-ieir15.technion.ac.il-16376-0 not empty

I was able to get rid of it by commenting out lines 75-78 in parallel.py,
but I'm not sure if this is the correct thing to do.