fix(zonos): add missing gradio launch command

Upstream Zyphra/Zonos ships no CMD in its Dockerfile (it launches the
app from its own compose), so our container ran the NVIDIA entrypoint,
printed the CUDA banner, exited 0, and restart-looped — nothing ever
bound 7860/8199. Add command: python3 gradio_interface.py to match
upstream, plus an explicit GRADIO_SHARE=False. Built + deployed to
irv-ml1; 8199 now serves HTTP 200 and the transformer model loads.

stacks/zonos/compose.yaml

@@ -29,6 +29,11 @@ services:
     container_name: zonos
     restart: unless-stopped
     runtime: nvidia
+    # Upstream's Dockerfile sets NO CMD — it launches the app from its own
+    # compose instead. Without this the NVIDIA entrypoint prints the CUDA
+    # banner and exits 0, looping forever (nothing binds 7860). Mirror
+    # upstream: python3 gradio_interface.py (demo.launch binds 0.0.0.0:7860).
+    command: ["python3", "gradio_interface.py"]
     ports:
       - "${ZONOS_BIND:-0.0.0.0}:${ZONOS_PORT}:7860"
     environment:
@@ -37,6 +42,10 @@ services:
       # Make Gradio bind all interfaces inside the container so the host
       # port-map reaches it (Gradio otherwise may bind 127.0.0.1 only).
       - GRADIO_SERVER_NAME=0.0.0.0
+      # Explicit: gradio_interface.py reads GRADIO_SHARE (defaults False).
+      # Pin it off so a missing default never tries a public share tunnel
+      # (irv-ml1 has no egress for that).
+      - GRADIO_SHARE=False
       - HF_HOME=/app/hf_cache
     volumes:
       - ${ZONOS_CACHE_DIR}:/app/hf_cache