MPI deadlock when MPI process is killed.

[Meijuh]

If one of the MPI processes is killed then using the current recommended settings for OpenMPI the mpi program will end up in a deadlock situation. This is wrong, if this happens then the whole application shoud terminate.

This is of course a bug in MPI rather than in LTSmin, but by adding deadlock detection we can make certain that we terminate.

Alfons: HREexit / HREabort facilities do not close down the comm channels in hre_pthread