Description of problem:
Please add support for socket activation in sshd, in the sense of of a listening
socket managed by systemd which is then passed to a long-running service on first
connection.

For many (most?) machines, sshd is used only occasionally. This includes both physical
hardware e.g. developer laptops, but also VMs/containers, where one might connect sometimes
for debugging, but not very often. It is good to have the possibility to connect, but always
starting sshd wastes some ram and cpu cycles at startup.

I expect we would want to have a permanently-activated sshd only on real multi-user
servers, or on worker nodes which accept many ssh connections. For a majority of
installations, socket activated sshd would be more appropriate.

We currently have sshd.socket with Accept=yes which starts per-connection instances
of sshd. Sshd is better at managing multiple connections than systemd, e.g. logs it much
better and handles failing connections better.

As a second-order feature, if socket activation is implemented, sshd could automatically
deactivate itself after a few minutes of idleness. Because systemd woudl still hold
the socket, there is no window where the socket is not listening and connections
would fail.

We have added socket activation to almost all services over the years, sshd is one
of the notable exceptions.