Add LD_LIBRARY_PATH=/app/.local/bin to container env for native llama.cpp binary

The manual native CUDA build's rpath points to /tmp/llama.cpp/build/bin/
(host build directory) instead of $ORIGIN. Setting LD_LIBRARY_PATH in the
Docker environment ensures the shared libraries are found inside the container
and survives container restarts (unlike the ephemeral ~/.config env file).

docker-compose.yml

@@ -21,6 +21,7 @@ services:
       # Ollama at http://host.docker.internal:11434.
       - "host.docker.internal:host-gateway"
     environment:
+      - LD_LIBRARY_PATH=/app/.local/bin
       - LLM_HOST=${LLM_HOST:-localhost}
       - LLM_HOSTS=${LLM_HOSTS:-}
       - OPENAI_API_KEY=${OPENAI_API_KEY:-}