Fix Tier 2 fallbacks: text-only chat model for reasoning, drop unsupported VLM provider call

DISCLAIMER §8: document auto-delete behaviour

UI: pin verdict at top with placeholder; auto-delete source+intermediates after audit (opt-out checkbox)

3-tier resilience: GPU retry -> HF Inference -> deterministic; add torchvision for VLM

Add torchvision (Qwen2-VL video sub-processor dep); switch default reasoning model to E4B (works on ZeroGPU CUDA emulation)

Pin kernels<0.15 to fix transformers hub_kernels LayerRepository error

Initial deploy: Document Integrity Verifier (LEGX)

initial commit