docs: document judge backend selection (llama_cpp local / transformers ZeroGPU on Space)

qwen call fixes, tried with LM studio server, hope it does not break anything :) (#4)

upodate Claude code and hf upload file (#2)