llama.cpp b10048 fixes Phi3, Bert, Plamo2/3, ChatGLM

The llama.cpp project has released build b10048, which includes a fix for Phi3, Bert, Plamo2/3, and ChatGLM models.

• Fixes issues with Phi3, Bert, Plamo2/3, and ChatGLM architectures.
• macOS Apple Silicon KleidiAI support is disabled in this build.
• openEuler builds are available for x86 and aarch64 architectures, including 910b ACL Graph support.
• Windows binaries now include CUDA 13.3 support alongside existing CUDA 12.4 options.
• Pre-built binaries are provided for Ubuntu (CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL), Android arm64, and macOS Intel.

This release provides updated binaries for various platforms and hardware accelerators to ensure compatibility with the specified model architectures.