gitlab-sshd: Recover from panics in any one SSH connection

I think that right now, a panic in a connection/channel goroutine for any one client connected to gitlab-sshd will terminate the whole server, because we don't recover() in, e.g.: handleConn

Analogously to net/http, we should protect ourselves from a panic handling any one connection (or channel?) and terminate just that connection/channel instead of taking down the whole server.