This causes a neighbours check to be rescheduled after we receive a hello
or IHU.

IHUs are now sent synchronously with hellos.  There's no longer an explicit
IHU interval, we send IHUs for all neighbours every three hellos, and for
marginal neighbours with each hello.

Oops... otherwise we crash.

This is now done in send_request.

This makes sure that any neighbour answering the request has up-to-date
information about us.

We only expire sources every 5 minutes, since it's a somewhat costly
operation.

Now flush_source itself checks for the source being flushable.