basho/riak_test#1363

@martincox - you may want to check this. It is to fix a nextgenrepl issue, but will change functionality in other systems. It is a very subtle change in the GET_FSM behaviour following GET.

Previously after completing a GET, only one action (other than nop) could be performed: read_repair or delete (i.e. trigger tombstone reaping). If the GET revealed that both need to occur (i.e. there needs to be a repair, and also the winning object is a tombstone so tombstone trigger also required), only the repair would be done.

This indirectly led to a weird issue, which could cause some serious problems.

Now, as well as read_repair and delete, an additional delete_repair option can occur in this circumstance. If delete_repair is triggered, the FSM will first repair then delete. So a fetch will always trigger delete when required.