Linux Vulkan + Auto GPU Detection - #11

Summary:

• New Vulkan inference engine (libllama-vulkan.so)
• Auto-detect GPU at startup; fallback to CPU if unavailable
• Linux note: install Vulkan loader (libvulkan.so.1) + vendor ICD

Behavior:

• engine=auto: try Vulkan → else CPU
• engine=llama-vulkan: force GPU
• engine=llama-cpu: force CPU

Linux requirements:

• Debian/Ubuntu: apt-get install -y libvulkan1 vulkan-tools
• NVIDIA: nvidia-vulkan-icd; AMD/Intel: mesa-vulkan-drivers

Testing:

• GPU machine: engine=auto → Vulkan in logs
• CPU-only: engine=auto → fallback to CPU with warning
• Force modes: engine=llama-vulkan / engine=llama-cpu

Packaging:

• Bundle libllama-vulkan.so
• Clear logs indicating selected engine