Consider the standard refund transaction setup. A transaction with a 2 of 2 output is committed to the block-chain that is spent by a refund transaction.
If the refund transaction has a locktime 2 years into the future, then it cannot be spent for at least two years.
On the one hand, the refund transaction is unconfirmed. But on the other hand, there is no risk of its input being double spent. Both parties are safe to assume that the transaction will eventually be included.
A hard fork which makes the refund transaction invalid effectively steals that output. At the absolute minimum, there should be a notice period, but it is better to just not have that problem in the first place.
If someone has a 1MB transaction that spends a 2 of 2 output but is locked for 5 years, is it fair to say to them that it is no longer spendable?
I am confused. In your example, it is in the blockchain and since you have the ability to spend it, then why would any fork make it so you cant spend it?
If you are saying there are some 1MB tx that has timelocked tx in the future that is already confirmed, I am not sure why that is relevant. Clearly all existing tx that are already confirmed would be grandfathered in.
So the limit on tx size (however it is done) would apply to post fork tx.
Sorry to be slow on this, but I dont see what type of unconfirmed tx we need to make sure it is valid post fork. If it requires to create a new spend that is less than a 1MB tx size, that doesnt lose funds, so I dont see the issue.