feat(tagger): Qwen-VL NL captions, pose extraction (YOLOv8-n), slide-panel model selector

feat(tagger): pose extraction + Qwen3-VL optional NL captioner + WD14x3 ensemble weights recalibration

feat(prod): rewrite pipeline + NoobAI + ensemble tagger (3xWD14+DeepDanbooru-ready) + 1000 artists + negative templates + history ext + scoring config