Add RewardHarness to reward and verifier libraries

[reacher-z]

Summary

• add RewardHarness to the LLM-as-judge / reward / verifier libraries section
• note its evolved scoring skills/tools, image-editing judgments, and GRPO-compatible reward
• keep the change to one annotated README entry

Checks

• searched the default branch and all-state issues/PRs for the title, arXiv ID, and canonical repository URL
• verified the paper and code links
• git diff --check passes

Disclosure

I am submitting this on behalf of paper co-author Yuxuan Zhang.