A2C training runs

Here are all the A2C training runs we have performed.

As a word of caution: None of these really worked, because we could not get the legal moves part working, resulting in the agent always losing.

We did try many implementations or variations to restrict the A2C agent to legal moves, but ultimately gave up due to time constraints.

This frustration ultimately led to using PPO, since there it was a bit easier for us to just use Stable Baselines MaskablePPO to restrict the agent to legal moves.

These runs simply serve as a reminder of the attempted work and the failure.