I banned people from public IRC channels on EFNet all the time. Google deplatforms SEO spam sites all the time. The internet is unusable without some kind of filter on participation.
In a past web forum I did infrastructure for, without sufficient moderation and constant updates to captchas/challenge questions, we would get over 25k signups a day (100 legitimate signups a day), and over 500k spam posts a day (<5k legit). You'd be sending every single client multiple-digit terabytes of literal spam, much of it illegal spam (CC theft dump sales, illegal online pharmacy, are common spam) and making them dig through it.
Imagine viewing a single thread, and downloading about 5GB a page of gzipped viagra ads covered in the entire text content of a Harry Potter book in a desperate attempt to get past some filter, and then having to filter through to get the one or two posts.
My entire problem with censorship, philosophically, is some unwanted third party making demands on my information input. The goals of moderation need not be at odds with my goal to determine for myself what I wish to read. Categorization and shared filter lists could go a long way here.
It seems like we want the same thing. If we don't have any kind of moderation, then that filtering is pushed out to each and every leaf node, instead of being handled centrally, and this is a huge demand on my attention for information input. Do you remember how bad email spam had gotten before we developed effective server-side spam filtering? Email had become basically unusable.
The categorization is handled upstream. The decision of what to do with that information is left to me. I think that's the ideal model.