[feat] VLM as judge for WM

[H1yori233]

Purpose

Adds a judge.third_person_separation metric to fastvideo.eval: a pairwise Gemini judge that, given two rollouts from the same first frame + control signal, picks the one that better separates the third-person character (foreground) from the background, and reports it as a candidate-vs-reference win-rate.

Checklist

• I ran pre-commit run --all-files and fixed all issues
• I added or updated tests for my changes
• I updated documentation if needed
• I considered GPU memory impact of my changes

[mergify]

Merge Protections

Your pull request matches the following merge protections and will not be merged until they are valid.

🔴 PR merge requirements

Waiting for

• #approved-reviews-by>=1
• check-success=fastcheck-passed
• check-success=full-suite-passed

This rule is failing.

• #approved-reviews-by>=1
• check-success=fastcheck-passed
• check-success=full-suite-passed
• check-success~=pre-commit
• title~=(?i)^\[(feat|feature|bugfix|fix|refactor|perf|ci|doc|docs|misc|chore|kernel|new.?model|skill|skills|infra)\]

[gemini-code-assist]

Code Review

This pull request introduces the judge.third_person_separation pairwise VLM metric using Gemini, along with an evaluation script, documentation, and dependency configurations. Feedback focuses on improving API robustness by falling back to a tie on failure, removing self.k from the cache path hash to enable proper cache reuse, using a deterministic hash for reproducible counterbalancing, and gracefully handling missing scores in the evaluation script output.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.