I think we are misunderstanding the effect of this change.
It's still "OK" for a 50k re-org to happen.
We're just saying that if it does, we will now have potentially introduced a hard fork between new client and old clients if the reorg contains earlier signaling for the most recent ISM soft fork and then blocks which do not conform to that soft fork before the block height encoded activation.
I think the argument is this doesn't substantially add to the confusion or usability of the system as its likely that old software won't even handle 50k block reorgs cleanly anyway and there will clearly have to be human coordination at the time of the event. In the unlikely event that the new chain does cause such a hard fork, that coordination can result in everyone upgrading to software that supports the new rules anyway.
So no, I don't think we should add a checkpoint. I think we should all just agree to a hard fork that only has a very very slim chance of any practical effect.
In response to Thomas' email. I think ideally we would treat these soft forks the way we did BIP30 which is to say that we're just introducing a further soft fork that applies to all blocks except for the historical exceptions. So then its a almost no-op soft fork with no risk of hard fork. This however isn't practical with at least BIP 34 without storing the hashes of all 200K blocks that don't meet the requirement.