CUDA FA: run KV_max mask scan for all Q batch sizes#22137

ssam18 wants to merge 1 commit intoggml-org:masterggml-org/llama.cpp:masterfrom

ssam18:fix/fattn-kv-max-small-batchssam18/llama.cpp:fix/fattn-kv-max-small-batchCopy head branch name to clipboard