fix(vlm): _clean_json returns first complete JSON object, not brace span

feat(vlm): Step 8c-ii — opencv deskew preprocessing for scanned images

feat(vlm): pivot to Qwen3-VL-8B-Instruct (bf16) off the broken FP8 path

fix(vlm): pin kernels<0.14 for deep-gemm FP8 metadata compatibility (attempt)

fix(vlm): add `kernels` for Qwen3.6 fine-grained FP8 inference

fix(spaces): add torchvision — Qwen3-VL processor hard-requires it

feat(spaces): Step 3 — VLM + Extract Space app.py, prompts shipped in core

docs+scaffold: collapse to 2 Spaces, adopt Qwen3.6/NuExtract3

feat: scaffold repo layout — Step 1 complete