Add upstream Qwen conditioning in original and runtime DCP formats, official H3 TAE, audio codec, tokenizer/processor and complete runtime asset manifest. Causal decoder now constructs the student directly without a native video VAE checkpoint. Remove the external-dependency chapter; preserve MiniMax-H3-only acknowledgements and all applicable upstream licenses. No trained renderer/decoder weight changes.
Stream-Render: release trained renderer and causal decoder (#1)
