No external binary is required now.

FluidAI uses the embedded `llama.xcframework` and calls `llama.cpp` C API directly.
