llama.cpp b9604 released with SYCL backend fix and multi-platform builds

The llama.cpp project, a high-performance C++ implementation for running large language models locally, has tagged version b9604. This release primarily addresses a broken CI build and release pipeline for the SYCL backend, which powers inference on Intel GPUs and other SYCL-compatible accelerators. The fix restores automated testing for both Ubuntu and Windows SYCL builds, removes stale cache artifacts, and ensures the `ccache` utility is properly configured to speed up subsequent builds. The update also disables several experimental backends (KleidiAI on macOS, SYCL FP32 on Linux, SYCL on Windows, and openEuler builds) due to ongoing stability work.

Beyond the SYCL correction, b9604 continues llama.cpp's commitment to broad platform support. The release ships pre-compiled binaries for macOS (Apple Silicon and Intel x64), Linux (x64 and ARM64 CPU builds, plus Vulkan, ROCm 7.2, and OpenVINO variants), Windows (x64 and ARM64 CPU, CUDA 12/13, Vulkan, and HIP for AMD GPUs), Android (ARM64), and iOS as an XCFramework. This allows developers and end-users to run quantized LLMs on everything from a Raspberry Pi to a multi-GPU workstation without manually compiling. The project, hosted under ggml-org on GitHub with over 116,000 stars and 19,500 forks, remains the most popular way to run models like Llama, Mistral, and Gemma offline.