llama.cpp b9957 improves server tools, removes apply_diff

The latest tag of llama.cpp, b9957, brings targeted enhancements to the server's tool-handling capabilities. The release removes the apply_diff function while introducing a new tools_io abstraction layer, alongside a basic implementation (tools_io_basic). This refactoring simplifies how developers integrate external tools with the llama.cpp inference server, making it easier to extend functionality without deep code modifications. The edit tool has also been improved, likely offering more robust support for code editing workflows.

This release is built for an extensive range of hardware and platforms, from macOS Apple Silicon with KleidiAI optimizations to Windows CUDA 12/13, Vulkan, and even openEuler. Developers can expect more predictable tool interactions and a cleaner API surface. The removal of apply_diff suggests a shift toward more atomic, modular tool operations. For teams running local LLMs for code generation, data analysis, or automation, b9957 offers a more maintainable foundation.