If you expose Jellyfin on 443, have HTTPS properly set up (which Caddy handles automatically), your admin password is not pswd1234 (or you straight up disable remote admin logins), and use a cheap .com domain rather than your IP--what is the actual attack surface in that case?

As far as I can remember that is more or less what is usually suggested by Jellyfin's devs, and I have yet to see something that convinces me about its inadequacy.

With a reverse proxy, I don't see how this would work. The whole way the reverse proxy works is you use a subdomain name ("jellyfin.yourdomain.org") to access Jellyfin, rather than some other service on your server. The reverse proxy sees the subdomain name that was used in the HTTP request, and routes the traffic based on that. Scanning only the IP address and port won't get attackers to Jellyfin; they need to know the subdomain name as well.