llama.cpp b9973 release with DeepseekV4 cache fix

The llama.cpp project has released version b9973, which includes a specific update for DeepseekV4 models. The primary change involves clearing the cache only for the sequence rather than the full context.

• macOS Apple Silicon and Intel binaries are available.
• iOS XCFramework is provided.
• Ubuntu builds cover CPU, Vulkan, ROCm 7.2, OpenVINO, and SYCL backends.
• Android arm64 CPU binary is included.
• Windows builds support CPU, OpenCL Adreno, CUDA 12/13, Vulkan, OpenVINO, SYCL, and HIP.
• openEuler x86 and aarch64 builds with ACL Graph are available.

This release provides updated binaries for various platforms and hardware accelerators.