llama.cpp

ggml-webgpu: parameterize submission size and add iOS specific limits

#21533

Merged