No one is going to waste the CPU time running Stockfish (or GPU with Leela) against each other at classical time controls long enough to see a decisive game which might never happen anyway so we will likely never know. I think current Stockfish on a modern ThreadRipper with classical time control will never be beaten. I am not 100% sure I am right but I don't expect (but would love to!) to be proven wrong.
Due to alpha-beta cutoffs, you don't need a full game tree. But if you did build a tree with all legal chess positions, it would have approximately 4.8x10^44 of them [1].
I am not saying Stockfish is going to solve it. I am saying it will not lose from the initial position. In other words my hypothesis is that we already have a soft solution to chess available. I can't prove it but I can proceed accordingly in practice (in opening preparation or correspondence games for example) and no one will prove me wrong or exploit it.
Stockfish lost a handful of games against Leela from starting positions that were a couple of moves into mainstream openings. I don't think it's at at all unlikely that there's a route that beats it.
Can you provide an example of such a game? I am following computer chess pretty closely and it's just all draws if we start from mainstream openings. By mainstream I mean things that are considered solid by today top players and engines. Some of the traditional mainstream openings turned out to be problematic (King's Indian for example). I am talking about recent Stockfish (the one with NNs) as the traditional version had pretty serious problems in certain positions and needed an extensive opening book to steer it out of trouble.
Not always. Almost always.