replace multiprocessing.Pool

[andreasvc]

multiprocessing pools work fine unless any kind of error condition arises...

• properly detect segmentation faults, out of memory, &c.
  concurrent.futures does this, but doesn't take an initializer ... (works on both python 3.3 and backported python 2 version)
• handle ctrl-C (resolved in python 3.3+ for both multiprocessing and concurrent.futures; cf. http://bugs.python.org/issue9205 )
• robustness: retry failed jobs (but this might make things worse in certain cases, so better to design jobs so that half-finished jobs can be resumed later)
• do not rely on copy-on-write memory inherited from parent process; load data in child processes.
• Full Python traceback from children. parser.workerfunc() decorator does this.