Guarantee hang analyzer collects core dumps for sharded clusters, at minimum

Attaching gdb and collecting diagnostics for all processes in a sharded cluster continues to time out after 15 minutes. BF-20581 is a recent example where only 6 of the 9 mongod processes were attached to. Server engineers may end up relying on good luck or having access to multiple occurrences to successfully interpret the cause of a hang.

We should consider 1. reordering the steps in hang analyzer so a core dump can be captured for every mongod process even if the diagnostics against the live process cannot, or 2. we should consider sending a SIGABRT to any process gcore wasn't run on before the 15 minutes expire.