feat(model): register minimax h3 as a text2image base model

Reference entries for the bf16 repo and the sdnq uint4 quant load the
modular pipeline through the standard dispatch. Image tabs run the
model in still mode with audio off; the video tab keeps its own
overrides through the shared per-generation hook. Detailer is not
supported and is disabled with a warning.

data/reference-base.json

@@ -374,6 +374,14 @@
     "size": 75.64,
     "date": "2025 September"
   },
+  "MiniMaxAI MiniMax-H3": {
+    "path": "MiniMaxAI/MiniMax-H3",
+    "preview": "MiniMaxAI--MiniMax-H3.jpg",
+    "desc": "MiniMax-H3 generates video with synchronized stereo audio in a single denoising pass through a 33B single-stream transformer with a Qwen3-VL conditioner. In image tabs the model runs in experimental still mode, keeping the first frame of a minimal generation.",
+    "extras": "sampler: None",
+    "size": 134,
+    "date": "2026 August"
+  },
   "Freepik F-Lite": {
     "path": "Freepik/F-Lite",
     "preview": "Freepik--F-Lite.jpg",