'Krasnol argued for keeping powerful/dangerous features, but making them opt-in (and a bit of a hassle to enable). You countered that there will be "some fraction" of users incapable of not hurting themselves with those features, who "will still stumble upon it anyways and will still refuse to take any responsibility for enabling it". My counter to that is that we shouldn't remove such power features just because "some fraction" may find and misuse them.

That's the should/should not part. The rest is my take on why companies remove those features anyway - they have no incentive to provide anything above bare minimum, especially not when they could be on the hook for "some fraction"'s mishaps.