drop llama-cpp-python (unbuildable on Spaces); explain MiMo-over-Gemma choice

build llama-cpp-python from sdist: abetlen wheels are musl-linked, Spaces is glibc

dual runtime: @spaces.GPU 4-bit NF4 path (Part B config) + llama.cpp CPU fallback

pin gradio sdk_version 5.50.0; drop gradio from requirements

MiMo-7B PDF injection detector: CPU/GGUF runtime, verbatim Part A+B logic