Add back Top-P sampling, Repetition Penalty, Logits Warpers

These are available in pre-1.0 releases, but were temporarily removed during the adaptation to `MLTensor` #257. Pull requests #23, #27, #85 can be used for reference.