Disallow reusing the same port unless explicitly requested?

[lilydjwg]

So that people can't accidentally start two instances on the same port and wonder why it sometimes works but sometimes gets wrong results.

The story: I had a bad time debugging a model repeating issue on a specific device but not another. It took me more than one hour, adjusting parameters and searching the web, until I finally decided to give up, stopped the instance, but while closing all the browser tabs, I hit the refresh button again, and was surprised there was still response! It turned out that I had a text embedding model running on the same port.