fix: resource leak? We're not using SetDeadline for UDP

[WofWca]

Does it mean that UDP "connections" are kept indefinitely? For the server it is potentially critical because it appears that just one client can exhaust all outgoing UDP ports of the server.
It's not about just memory because there are only 65535 ports, and we can probably handle that many pieces of state, memory-wise.

Is it the same for TCP?

But even with SetDeadline, there is nothing stopping the client from just keeping the connections alive?

I guess this is similar to how SOCKS proxies work? They also face a lot of parallel TCP connections, per client.

Adding Prometheus metrics would be nice.