TLS is widely supported with dozens of available implementations ready-to-use in many different programming languages and on many different platforms, and it basically comes free for any browser-based implementation. Those implementations also receive a lot of attention, and because of that library support it is much easier to update an application that uses TLS than some purpose-built chat protocol. For example, let's say a new EC attack is discovered and we have to move everyone to a different set of curves (e.g. maybe P256 is found to be insecure and we all have to switch to P521). An OpenSSL update will be pushed out a lot sooner, and will be used by far more client applications, than the updates to all of the hundreds of chat clients that need whatever chat-specific e2ee protocol updated.

At the end of the day, even with all problems that exist in TLS implementations, I have a lot more faith in TLS than I do in some college student's hacked together web chat client's e2ee implementation. As for XMPP, just how widely available is OMEMO in XMPP client software? The last time I tried to deal with XMPP and e2ee I was constantly confronted with clients that did not support this or that protocol. I can't speak for Matrix, maybe it "solved" the problem, but as I said if e2ee was not part of the standard from the beginning it is going to be hard to push it out as an afterthought.