llama.cpp b9259 fixes speculative decoding nullptr crash for stability

The llama.cpp project, a widely-used C/C++ implementation of LLM inference, has issued release b9259 to address a critical crash in speculative decoding. The bug occurred in the `get_devices_str` helper function, where `ggml_backend_dev_by_name` always appends a null pointer sentinel to the device vector. Earlier code did not skip these null entries when calling `ggml_backend_dev_name`, leading to assertion failures and crashes on certain system configurations. The fix ensures null pointers are properly skipped, restoring stability for users leveraging speculative decoding—a technique that accelerates text generation by using a smaller draft model to propose tokens while the larger target model verifies them in parallel.

This maintenance release also reaffirms llama.cpp’s commitment to broad platform support. Pre-built binaries are available for macOS (both Apple Silicon and Intel), iOS, multiple Linux distributions (x64 and arm64 with backends including Vulkan, ROCm 7.2, OpenVINO, and SYCL), Windows (x64 and arm64 with CUDA 12/13, Vulkan, SYCL, and HIP), Android arm64, and openEuler. The fix is already merged into the main branch, and users running previous versions are encouraged to upgrade to avoid speculative decoding crashes. This update is particularly relevant for developers deploying local LLM inference pipelines and for anyone using llama.cpp in production or edge environments where reliability is paramount.