CUDA: fix race conditions in FlashAttention kernels#13438

JohannesGaessler merged 1 commit intoggml-org:masterggml-org/llama.cpp:masterfrom

JohannesGaessler:cuda-fa-fix-race-condJohannesGaessler/llama.cpp:cuda-fa-fix-race-condCopy head branch name to clipboard