llama.cpp b11046 adds OpenCL flash attention

llama.cpp has released version b11046, introducing OpenCL support for the flash_attn_f32_f16_bin kernel and guarded prefill optimizations. The release maintains broad compatibility with existing platforms including CUDA 12/13, ROCm 10.0, Vulkan, and various CPU architectures across macOS, Linux, Windows, and Android. This update specifically targets OpenCL users, allowing them to leverage flash attention mechanisms previously limited to other backends. The build includes standard binaries for major operating systems and hardware configurations.