fix: ep 16 & 17 from the test set had a non-equal number of steps, we removed the extra step from ep16, whilst imperfect these now align

initial commit of v2 dataset, this has a dedicated training set with each ODD EPISODE ID from the test set

initial commit