llama.cpp b9891 adds col2im_1d Metal op and releases binaries

The llama.cpp project released version b9891, introducing a new `col2im_1d` operation for the Metal backend. This addition mirrors the CPU and CUDA paths, supporting f32, f16, and bf16 data types.

• The Metal kernel implements a gather operation with an F32 accumulator, linear indexing, and no atomics.
• Support checks were added to ensure destination contiguity and type matching for `GGML_OP_COL2IM_1D`.
• Binaries are available for macOS (Apple Silicon and Intel), iOS, Linux (CPU, Vulkan, ROCm, OpenVINO, SYCL), Android, Windows (CPU, OpenCL, CUDA 12/13, Vulkan, OpenVINO, SYCL, HIP), and openEuler.

This release enables the use of the `col2im_1d` operation on Apple Silicon devices via Metal.