Remove dead chat code and obsolete planning docs.

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

Route ZeroGPU auto backend to Transformers

Fall back from Transformers to llama.cpp

Select extraction backend by CUDA availability