The plug-in no longer calls external LLM

Key features

• The first three stages are LLM stages: inside DSH the session model itself does them — the model you're chatting with (Doubao, DeepSeek, whatever) is the brain that writes the story, script and shots
• Final cut with optional lip-sync stage: with a capabilities.lipSync provider configured, every voiced shot runs audio-driven lip sync (video reference + audio base64) and the synced clip replaces the original on the timeline
• Host routes/dsh-video-studio/health, /dsh-video-studio/runs, /dsh-video-studio/accounts (GET list masked / POST add / DELETE remove)
• the plugin never calls an external LLM.
• Mapping to the industry-standard toolchain