CUDA: Optimize `reduce_rows_f32` kernel, leading up to 25x perf improvement on kernel-level and 10% perf increase for Gemma3n#15132

JohannesGaessler merged 15 commits intoggml-org:masterggml-org/llama.cpp:masterfrom

ORippler:osimons/optimize_reduce_rows_f32ORippler/llama.cpp:osimons/optimize_reduce_rows_f32Copy head branch name to clipboard