llama.cpp b11044 updates Hexagon and CUDA builds

llama.cpp has released version b11044, focusing on backend optimizations for Hexagon NPUs and expanded CUDA support. The update refines IM2COL operations for Qualcomm's Hexagon architecture, improving efficiency for vision-based workloads on mobile hardware. Additionally, the release includes binaries for CUDA 12.8 and 13.3 on Linux and Windows, alongside ROCm 10.0 builds. This maintains llama.cpp's position as a versatile inference engine capable of leveraging diverse hardware accelerators.