If your client application asks for streaming and you see polling in action,
this must be because something in your network infrastructure is not compatible with streaming and the Stream-sense mechanism comes into play.
By looking at the Server log, we could confirm that.

Note that Allegro version has a built-in maximum frequency limit of one update per second on each item. Is it possible that you are confusing that with polling?

When the client loses the connection and reconnects, it is not bound to sending the messages occurred during the interruption. But see this FAQ for a broad discussion.
By the way, during the interruption can you see unsubscribe requests for the various items reaching your socket-based Remote Server, followed by new subscribe requests?