llama.cpp backend (16 KB page-aligned JNI native libs) for DVAI Bridge on Android. Provides GGUF model loading + inference; depends on android-shared-core for the HTTP/handler layer.