MiniMax H3 adaLN Modality

Scale a stacked LoRA's adaLN modulation per modality. H3 runs audio and video through the same 50 blocks, so there are no audio layers to target the way LTX 2.3 allows. The one pathway that does separate cleanly is adaLN: its projection emits three contiguous row blocks, one per modality, so scaling a slice steers that modality's modulation exactly. Wire the output into the stack's ``adaln_modality`` input. Leaving all three at 1.0 is a no-op; 0.0 removes that modality's share of the adapter. Only affects LoRAs that carry adaLN pairs -- the stack's report says, per LoRA, whether the control was actually live. It does not isolate a modality: attention is joint over the packed sequence, so this changes where a LoRA is applied, not everything it eventually reaches.