llama.cpp b9756 fixes server crash on file append overflow

The latest release of llama.cpp, version b9756, addresses a specific but impactful bug in the server's edit_file function. The crash occurred when appending content to the end of a file, triggered by a line_start value of -1. This value was incorrectly normalized to n+1, causing an out-of-bounds insert beyond the vector's end() iterator, leading to a heap-buffer-overflow. The fix normalizes -1 to n (insert at the last position), restricts -1 to append mode, and rejects it for replace/delete operations. Additionally, the parenthesization of the insert offset ensures that empty-file appends compute the position as int first, avoiding a transient begin() - 1 on a null vector data pointer.

Alongside this fix, the release continues llama.cpp's comprehensive cross-platform support. Build artifacts are provided for Apple Silicon (arm64, including a KleidiAI-enabled variant), Intel macOS, iOS XCFramework, Ubuntu x64/arm64 with CPU, Vulkan, ROCm, OpenVINO, SYCL, Android arm64, and Windows x64/arm64 with CUDA, Vulkan, OpenCL, and HIP. Notably, the KleidiAI-enabled macOS build is marked as DISABLED in the release notes, suggesting it may be experimental. This release is a maintenance update focused on stability rather than new features, but it is critical for anyone running the server with file editing enabled.