llama.cpp b9898 release fixes AOT double type issue

The llama.cpp project has released version b9898, which includes a fix for an Ahead-Of-Time (AOT) compilation double type issue via SYCL functions.

This update provides binaries for macOS (Apple Silicon and Intel), iOS, Linux (Ubuntu x64, arm64, s390x), Android, Windows, and openEuler across CPU, Vulkan, ROCm, OpenVINO, SYCL, CUDA, and HIP backends. Notably, KleidiAI support on macOS Apple Silicon is disabled in this release.

The release ensures compatibility for users relying on specific hardware accelerators and operating systems by addressing the identified compilation error.