Dmitry and Konstantin, I think I've restored C-Reduce's old behavior of
eagerly deleting temp dirs.
A side effect of this fix is that parallel execution may become more
efficient since C-Reduce is now killing off entire speculative process
groups instead of just killing its direct children.
Also, I made caching of the delta tests work again. It's not clear
whether this is worth it; measurements will be sort of painful since the
benefit is only going to matter on those multiple-hour reductions for
big C++ programs.
Anyway, let me know how the current version works if you get a chance.
John
