Fix cases where parallel execution gets stuck.

[mbj]

When running mutant multiple times over the same configuration it gets stuck in 1 of 10 runs. Killing with SIGINT is still possible. Reducing the amount of concurrent jobs seems to fix or reduce the chance of a problem.

Initial research shows that in some cases a segfault in a kill fork does not get signaled properly to the parent, leaving the parent waiting infinitely for a result.

This is also partially related to the timeout question raised in: #87.