I am disagreeing because I regard what you call "risk of escalation to assault" (or, more generally: risks of bodily harm and benefits from tightly-knit social cooperation) as a critical determinant of social interaction. It is very hard to meaningfully compare real-world and online contexts, much less regard them as "the exact same scenario".

(Indeed, I have jokingly argued before that we should totally deprecate and taboo the term "community" as referring to online social groups, since it tends to promote this very kind of ontological confusion.)

As for your question about "toxicity", let's just say that this particular discussion has been held already. If anything, LW has seemed to err towards taking complaints about divisive or disruptive behavior more seriously than they otherwise would, especially when outgroup status is a factor.