llama.cpp b9535 release fixes LFM2/LFM2.5 tool parser for AI agents

ggml-org's llama.cpp, the widely-used open-source C++ library for running large language models locally, has released version b9535. This maintenance release focuses on a critical fix for the LFM2 and LFM2.5 tool parser, which is used when models need to output structured tool calls (e.g., function calling for AI agents). The commit message reads: "common/chat : unify and fix LFM2/LFM2.5 tool parser (#24178)." This unification ensures that tool parsing behaves consistently across different model formats and chat interfaces, preventing errors when LLMs try to use external tools like web search, APIs, or file operations.

The release comes with extensive platform support, including builds for macOS (Apple Silicon arm64 with optional KleidiAI acceleration, Intel x64), Windows (CPU, CUDA 12/13, Vulkan, HIP), Linux (x64, arm64, s390x with Vulkan, ROCm 7.2, OpenVINO, SYCL FP32), iOS as XCFramework, and Android arm64. Notably, some configs like KleidiAI and SYCL are disabled in this release. The build matrix shows the team's commitment to broad compatibility, from consumer CPUs to enterprise GPU setups. For developers running local AI agents, this update is a quality-of-life improvement that makes llama.cpp more reliable for tool-using workflows.