Select llama.cpp GGUF on CPU Basic Spaces

Expand lab knowledge graph to 107 markers with curated YouTube videos.

Add agent trace panel, vision extraction, and local Transformers backend.

Route ZeroGPU auto backend to Transformers

Select extraction backend by CUDA availability

Switch Space deployment to ZeroGPU

feat(extraction): GBNF grammar + local MiniCPM-V (llama.cpp) backend + factory