Multi-writes in a sharded cluster (updateMany:true and justOne:false for deletes) do not perform version checking on account that they are broadcast to all nodes in the sharded cluster. Such operations attach the special value ChunkVersion::IGNORED to indicate that an operation is coming from a router (as opposed to direct connection to a shard), but that the shard must not perform version checking, under the assumption that the caller knows what they are doing.

However, ChunkVersion::IGNORED still triggers a StaleShardVersion exception in the case where the shardVersion is UNKNOWN or if the shard is in a critical section.

The former is not a big problem, since it only happens once for the duration of a shard's MongoD process, but the latter is problematic since it may exhaust the 10 retry attempts that we allow on the router.

This ticket is to come-up with a scheme so that multi-writes' StaleShardVersion exceptions be retried at the level of the shard and not bubble up all the way up to the router.