When we start recovery of a shard we should wipe the state file of the copy if it's present otherwise gateway allocating can get confused interpreting a shard that is not fully recovered ie. due to a recovery failure as a valid copy since we only write the state when the shard is started.