Add ExtractArena tab to the Space app

Force sdpa attention on CUDA to stop hunyuan-ocr OOMs

Guard against oversized image uploads OOMing the council

adding manual requirements

Convert svlm_council into the ZeroGPU Space app (Gradio + MCP)

Swap Infinity-Parser2 for Muse-Glimmer-30B + two Gemma 4s in ExtractArena

Add council-models-refresh skill for the recurring model-update workflow

Document model research in MODEL_RESEARCH.md, link from README/CLAUDE.md

Add sVLM-Council: local small-VLM arena, wired into ExtractArena

Add optional --format/-F output-format directive to ExtractArena

Guard ExtractArena prompts against hallucinated field values

Add ExtractArena CLI for field extraction across three VLMs

Add OCR arena CLI comparing moondream3, Qwen3.5-9B, and GLM-OCR

Initial commit