Should we fail engine immediately after a delete fails? #33256
Description
This is a spin-off of #33217 (comment).
Today when a delete operation on a primary fails, we won't immediately fails that shard but "replace" that delete by a NoOp. While this approach sounds correct, its benefit is questionable. What can we recover after a delete fails? If nothing, should we just fail a shard immediately?
Let's revisit this approach.