Deploy RustLean v4 GGUF

Install nvidia-cuda-runtime-cu12 + nvidia-cublas-cu12 so libllama.so finds libcudart/libcublas

Lower ZeroGPU duration to 45s (model pre-downloaded; cuts quota per call)

Surface ZeroGPU worker traceback via try/except for diagnosis

Preload nvidia CUDA libs (RTLD_GLOBAL) so libllama.so finds libcudart on ZeroGPU; eager model download; duration=120

gradio 6: Examples nested list, Slider keyword args

gradio 6 fixes: Code language=None (rust unsupported), move theme to launch()

Defer llama_cpp import into @spaces.GPU path (ZeroGPU: CUDA libs only exist inside the GPU call)

Fix ZeroGPU hook: restore gradio 6.20.0 (matches SDK), drop gradio from requirements

RustLean FIM Space: cursor-aware Rust completion via llama-cpp-python + ZeroGPU