feat(video): add Wan 2.2 and LTX 2.5 CUDA rendering

[atomantic]

Summary

• add cache-only, pinned CUDA runtimes for Wan 2.2 TI2V and LTX 2.5
• add setup/repair flows, model registry metadata, disclosure, and streamed runner dispatch
• require Python for mixed-runtime installs
• enforce LTX image/source pairing and cap Wan CUDA to text mode at both request and render boundaries

Testing

• focused server/runtime suite: 385 passed, 13 skipped
• bash -n scripts/setup-image-video.sh
• Python AST validation was unavailable because this Windows host''s configured Python launcher target is missing; runner dispatch and argument construction are covered by the Node test suite