fix: gunicorn start command for production server port

[emidiovaleretto]

What

Replaces the Dockerfile start command with gunicorn (shell form) and
removes the dev runserver command.

Why

The production deploy was returning 502. The deploy logs showed:
Error: '$PORT' is not a valid port number.

Two problems were behind it:

1. The container was still starting with Django's runserver (dev server),
   not gunicorn.
2. The gunicorn command was being passed in a way that left $PORT
   unexpanded — the literal string $PORT was reaching gunicorn instead
   of the port number Railway injects at runtime.

Using the shell form of CMD lets the shell expand $PORT to the real
port, so gunicorn binds correctly.

How to verify

• Deploy logs no longer show the $PORT error and gunicorn starts
• The app responds at the Railway URL (no more 502)
• Migrations run via the release step in the Procfile