feat(kore): ts interpreter fixes

[paradite]

Some issues surfaced when I started actual training, so here are the fixes.

@bovard kindly help review please.

So far the progress has been good with the new interpreter. Simulation is actually much faster than using kaggle-environments python cli. I will share the new DQN baseline setup soon on discussion forum.

[paradite]

I have encountered a new issue:
while the model does well against the starter bot / old agent in new step interpreter written in TS (during training and after-training evaluation), it is not performing as well in the python interpreter.

I suspect there are more major issues with the TS interpreter or the Board.next() implementation. Hence I am reverting this back to draft and taking a closer look.