I reverted the change because as mentioned, to properly protect minority-led fork, the check need to be done by all applications interested in chainID based replay-protection. It suffice one important application to be put the minority-led fork at a disadvantage.
So now EIP-1959 assume that no protection will be put in place for minority-led hardfork. I added some info in the rationale.
Instead I also created a new EIP-1965 that deal with the issue. It would be the solution I would vote for.
Discussion here : EIP-1965 Valid ChainID For Specific BlockNumber : protect all forks