The checkShardPrecondition phase of addShard before we check the replicaset cleanness we block user writes and if the replicaset had any data, then we unblock the user writes and fail. this is not 100% safe as the RS might had user write blocking already before we set it up, and in this case a failing addshard practically just unlocks the RS.

the idea is to store the original blocking phase in the add shard coordinator document and use that data on failure.