fix: add Podman support with NVIDIA CDI GPU passthrough

[mariojcr]

Problem

The project assumes Docker as the only container engine. Podman users hit two issues:

1. GPU passthrough fails — The deploy.resources.reservations.devices syntax is Docker-specific. Podman uses CDI (Container Device Interface) instead.
2. Image pulls hang or fail — Short image names like redis:alpine or ollama/ollama trigger Podman's interactive registry prompt (or fail in non-interactive mode), since Podman doesn't default to docker.io.

Changes

docker-compose.yaml

• Document both GPU options: Docker (deploy block) and Podman (devices: nvidia.com/gpu=all)
• Fully qualify all Docker Hub images with docker.io/ prefix

scripts/install.sh

• Auto-detect Podman or Docker at install time
• Handle podman compose and podman-compose as compose backends
• Validate NVIDIA GPU via CDI specs when running under Podman
• Generate the correct GPU syntax in the compose file based on the detected engine
• Fully qualify all Docker Hub image references

Makefile

• Add CONTAINER_ENGINE variable (auto-detects podman, falls back to docker)
• Fully qualify image references

Notes

• No behavioral change for Docker users — all defaults remain the same
• Podman GPU requires the NVIDIA Container Toolkit with CDI specs generated:

  sudo nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml