Add a not-so-happy path for the deletion of shards in ingesters upon source/index deletion

Follow up on #4244 

If the attempt to remove the shard in the ingester fails (ingerster temporarily unavailable for isntance)
we need a way to reliably make sure they eventually get the message.

Above 1mn or so, the lack of truncate message for the shards will block mrecordlog.

