feat(plugins): register LongCat-Video-Avatar 1.5 plugin

[tong-io]

Summary

• Registers tongflow-modal-longcat-avatar: LongCat-Video-Avatar 1.5 (Meituan, MIT) for the audio-image-gen-video slot, as an alternative to InfiniteTalk (same team's successor line: Whisper-large-v3 audio encoder, 8-step DMD distillation).
• Adds it to config/official-plugins.json (after InfiniteTalk), the GPU plugin lists of all three READMEs, and its icon.
• Capability matrix unchanged: the slot already has InfiniteTalk, which stays the default.

Plugin notes

• Single H100, bf16 DiT + DMD LoRA, no torchrun / context parallelism (all CP calls in upstream are gated on cp_size > 1).
• Long audio is covered by video continuation segments (3.7 s first clip + 3.2 s each), capped at 60 s.
• ~60 GB of weights from two public HF repos; no HF token needed.

Test plan

• tongflow.scan picks up the plugin with no errors
• modal run download.py::download, modal deploy deploy.py, real run on a portrait + speech clip

🤖 Generated with Claude Code