fix(llama-cpp): populate tensor_buft_override buffer so llama-cpp properly performs fit calculations

fix(llama-cpp): populate tensor_buft_override buffer so llama-cpp properly performs fit calculations #10561