llama.cpp b10061 initializes tensors in test_dsv4_hc to avoid NaNs

The llama.cpp project released build b10061, which includes a fix for the `test_dsv4_hc` test suite. The change initializes all tensors within that specific test to prevent NaN values from appearing in sentinel tensors.

This release provides binaries for macOS (Apple Silicon and Intel), iOS, Linux (Ubuntu with CPU, Vulkan, ROCm, OpenVINO, and SYCL backends), Android, Windows (CPU, CUDA 12/13, Vulkan, OpenCL, OpenVINO, SYCL, HIP), and openEuler.

The update ensures test stability by addressing the NaN issue in the specified test case.