Well, presenting some of those arguments in 2023 is rather embarassing.

There are watchtowers. And the timeout for big channels (>10M sats I think) is about a week by default, so even if not using watchtower, it should be enough to bring your wallet online once a week, far from 100% uptime. And this is only needed if you receive by lightning. If you are only paying, your past balance is always greater or equal than your current balance, so your channel peer has nothing to gain by broadcasting the old channel state.

But Lightning allows you to "compress" - you only need 2 onchain txs (opening and closing the channel) instead of N, where N can in principle be any number.

There are some true statements - if you receive on-chain, you have to spend on-chain. But how is this a problem of lightning? And if you are spending on-chain, you may instead open a channel for comparable cost.

Lightning doesn't solve everything, but it is a part of scaling. There are other ideas, like Ark and Enigma.