Add RewardBench 2 result: 25.3 average (random floor) with full protocol - the OOD boundary measured on a third-party benchmark

Fix usage snippet: pairwise comparison, not absolute score; add chance floor + public baseline rows

Upload README.md with huggingface_hub

Upload DebertaV2ForSequenceClassification