llama.cpp found its Vulkan flash-attention dequant path only engaged when the KV cache was full.

Build b10781 fixes PR #28190: the FA dequant path skipped engagement because of an nb[3] stride check the shader never reads when ne[3] == 1. Cache views carry the full-buffer stride there, so the check collapsed to n_kv == kv_size and the optimized path only turned on with the cache completely full. Anyone who benchmarked Vulkan flash attention on llama.cpp before this build was measuring the unoptimized path for nearly every real workload. Build b10782 separately unblocks CUDA graph optimization for multi-GPU under GGML_CUDA_GRAPH_OPT=1.