The agents have their utility and their endowments drawn from a uniform over the unit interval, they max their utility to determine their demand and observe the price when it is their turn to trade. The CFMM has initial reserves (900,1111.11), they show steady state behaviour after 25,000 trades.
Yes, it is stronger assumption than that: that the amount they can trade in the other venues at that price is much bigger than the CFMM amounts.
They would break the WE prices, so don’t fit quite well in the welfare analysis. They are a hard problem and out of scope for this work I would say.
When there is congestion (more transactions than blockspace to fit them all in) there is some MEV in this formal definition, from the fact that something will get excluded and picking what is valuable.