Echo chambers were created by people censoring viewpoints within their networks. The solution to that problem is not more censorship.
The web is a complex system that is more than capable of producing outcomes that are drastically different from the sum of its inputs (i.e. "people"). Based on what kinds of mechanism design and incentive structures are put in place, it is not only probable, but more mathematically likely that even a web filled only with rational correspondents will eventually devolve into a mess filled with falsehoods and echo chambers.
The important corollary to this insight? There's an entire class of structural remedies we can take that you're not even considering, before we even approach anything remotely resembling censorship on the individual.
It was slower than Twitter/Facebook but the effect is exactly the same as the social media problem. I have a relative who went down the conservative conspiracy theory rabbit hole in the years after 9/11 and the web was a huge accelerant for his intellectual decline because once he found a community of fellow travelers he’d stay in that network of blogs where they strenuously resisted outside sources or the general idea of objective reality. When those communities moved primarily onto social media, they moved faster and got wider range but the practices didn’t change at all.
I agree with you insofar as Mill's free market of ideas is the best remedy against bigotry and extremism, but you have the wrong idea about how to achieve that -- unregulated laissez-faire "freedom" on the web does not result in actual free discourse. On the contrary, that's just a recipe for a web overrun by all manner of adware, malware, spyware, propagation of unconscious self-censorship by means of universal targeted and mass surveillance, and straight-up disinformation ops and public opinion manipulation from hostile actors. How are you going to have sincere exchanges of ideas at scale over all that noise?
Similarly, there’s not that big difference on “policing” - the social media companies enforce very basic terms of service but most people use web hosts with relatively similar terms. The major difference is the discovery mechanism (and the fact checking applied to certain politicians’ accounts which aren’t banned on public interest grounds) which is closer to how search engines avoid sending links to certain sites with bad reputations. A lot of what Twitter/Facebook block will also be dropped by Google or many free web hosts. You can definitely get away with more but I don’t see much reason to believe that it’s a major difference rather than minor variations depending on which companies and content you’re talking about.
That by no means seems to be a sure assumption today. If anything, the question now is whether the latter state of affairs might not be the default, and if so, whether it is reversible at all.