CUDA: Enable cuda graphs for qwen3 next-style architectures#19521

ORippler wants to merge 3 commits intoggml-org:masterggml-org/llama.cpp:masterfrom

ORippler:osimons/cuda_graphs_qwen3_nextORippler/llama.cpp:osimons/cuda_graphs_qwen3_nextCopy head branch name to clipboard