ggml: fix cuda kernel launch configuration for k_compute_batched_ptrs to support large batch#16744

JohannesGaessler merged 4 commits intoggml-org:masterggml-org/llama.cpp:masterfrom

leejet:cuda_mul_mat_fixleejet/llama.cpp:cuda_mul_mat_fixCopy head branch name to clipboard