I think the most important is preventing HTTP connections, because private websites are most likely to be exploited via HTTP requests. Preventing any web traffic is probably a tad bit less important, but still useful - one can imagine buggy servers mishandling malicious malformed data, or even vulnerable websocket server implementations.

I have not yet formed an opinion regarding WebRTC and WebTransport.

WebSocket is essentially an HTTP connection, so it would have to be covered.

WebTransport uses ALPN so maybe that is fine indeed. We would need to make it very clear that not all browser connections to these network addresses are preflighted.

Unsure about WebRTC as well, but looking into it.

Given that we had to expand port blocking to WebRTC, it would make sense to me that this also applies there.