Add README.md: verified eval wrapper + fix guide for eval harness

Add robodyna_policy.py: verified eval wrapper + fix guide for eval harness

Add EVAL_FIX.md: verified eval wrapper + fix guide for eval harness

Fix eval I/O: images are CHW (3,H,W) not HWC; require inference repack. Verified model reproduces training actions (MAE ~0.01, corr ~0.9998).

Fix robot identity: dual-arm UR5 (was mislabeled aloha-agilex)

Add pi0.5 RoboDyna checkpoint (step 22000) + eval README

Upload README.md with huggingface_hub

initial commit