The original problem is symmetrical: there is a potential trade which will benefit both A and B, and they need to strike a price. The Ultimatum game is asymmetrical: one player goes first. This seems to me a conclusive proof that this problem cannot be modelled as an Ultimatum game.

You can see any problem as one-step by deciding a whole (possibly infinite) strategy instead of just the next action.

I don't think this works in the large unless P=NP (or something of the sort). In the small, e.g. analysing chess, it reduces the problem to no steps at all: both players exhaustively analyse the game and know the outcome without playing a single move. (I'm using "small" and "large" in the sense of the dispute between small-world and large-world Bayesians.) If that worked for the bargaining problem, A and B would independently come up with the same price and no bargaining process would be necessary. No-one has posted a method of doing so.