This work is being split out of SERVER-26016.

Like our existing replica set passthroughs, but randomly sends kill -9 to the (non-voting) secondary. It should be possible to start the secondary without connectivity to the primary (using mongobridge) and have it reach the SECONDARY state quickly. When connectivity is restored, it should be able to catch up to the primary and pass our repl validation tests.

Optional: It may be worth starting without "--replSet" before it replays the oplog to test the invariants (begin <= minValid <= oplogDeletePoint <= top of oplog). Any NULL items should be removed from the comparison, but transitive orderings still hold (minvalid <= top of oplog, even if oplogDeletePoint is null).

Optional: It may be worth pausing replication for most of the run, then unpausing before killing to maximize the probability of killing when not idle. (see clean_shutdown_oplog_state.js for an example of this technique)