llama.cpp b10249 adds get_info tool for server diagnostics

The llama.cpp project, known for running large language models efficiently on consumer hardware, shipped tag b10249 on August 4, 2025. The standout feature is a new server-side get_info tool (PR #26522), which lets connected clients query the host system's details, such as operating system and working directory. The implementation was carefully designed to handle probe process failures gracefully—if the probe can't spawn or times out, the server reports the OS as 'unknown' and never leaks diagnostic text to clients.

In addition to the get_info tool, the release fixes a --rpc documentation issue and simplifies result handling by consolidating limit logic and removing error branches. The release includes prebuilt binaries for a wide range of platforms: macOS (Apple Silicon and Intel), iOS, Linux (CPU, Vulkan, ROCm, OpenVINO, SYCL), Android (arm64), Windows (CPU, CUDA 12/13, Vulkan, OpenCL Adreno), and openEuler. Notably, the macOS Apple Silicon build comes with a KleidiAI-enabled variant, and s390x Linux is supported. This update continues llama.cpp's pattern of regular, incremental improvements to its local LLM serving stack.