The obvious counter to the notion that the issues with Mastodon can't be fixed is that nothing would prevent a carbon-copy of Twitter (or Twitter itself, for that matter) to federate. It's
not about Mastodon, but about ActivityPub.
And sure, most users will end up centralised on a handful of servers, and that is not a problem as long as they can move when a specific server does not meet their needs. The biggest importance is a safety valve.
I don't know which parts of the UI you think can't be fixed, but I don't agree. Better search is needed, but can be externalised (nothing stops someone from federating with the main instances and signing up to a few relays and offering more comprehensive search). A better experience when looking up external users timelines is largely down to Mastodon's developers having very specific ideas about how to do things. The logic for following or boosting/liking toots when you're on a different instance's timeline is awkward but can be improved with an optional intermediary (falling back to the awkward request for your address for those who insist on opting out of an intermediary providing a mapping).
The upside with Mastodon is that it's not a single thing. It's the server, and the web UI, and you can replace either or both. E.g. you can run Pleroma with the Mastodon web UI, or it's own, and there are additional web UI's. Or you can (and people do, and I'm doing so to) build your own, so we'll see a number of choices of platforms that will look very differently.
Personally I hope Elon fails, because it will make things easier, but I think the long term belongs to ActivityPub over Twitter in any case for the simple reason that the presence of ActivityPub means that the threshold to stand up a viable social network has dropped massively. On one hand you have the software you can set up in less than a day (been there, done that), but far more importantly, sign up to a relay and within hours you have an instance that looks lively, even though most toots are via federation. Where there before was a massive chicken and egg scenario, there's now an increasingly vibrant starting point for people to experiment with their own visions for what social media should be, and each one of them will add to the network effects of ActivityPub.
Mastodon may well fail, but social networks has a history of reaching tipping points that very abruptly changes things dramatically. In both directions. But the federated networks are far less vulnerable to people moving elsewhere.
As for the failures of federation, e-mail and mailing lists is the obvious counter. As is the web. Social media is a newfangled baby in comparison to the successful federated services.