__fenix_repair_ranks could hang due to permanent failure to repair communicators

[rfvander]

The whole program could hang if success of some of the repair operations remains elusive (endless loop over “while (!repair_success) {“). I think I remarked earlier that there was a stray variable num_try that gets incremented but not used that by function. Evidently, it was introduced to set some kind of limit on attempts to repair communicators. I think we should introduce an environment variable to govern the number of retries, with a default value set to a reasonably large number.