Yes, but all these solutions require the optimizer to be smart enough to be smart enough to optimize them away: "oh, owner.id is the primary key, so any value from owner will appear at most once, I don't have to group by them or track the maximum value, I can pick and store any value".

I think all optimizers are smart enough already, as this is a super common pattern. I honestly would have preferred this logic to be made more explicit: if you group by the primary key (or by the table name, as someone suggested), you can use all columns from the table without aggregate functions.

This kinda breaks down when you introduce subqueries and CTEs, which is where myopic optimizers can screw the pooch as well, so any_value can be of use there.