This avoids desynchronisation between Babel and the kernel.

This fixes a (minor) bug that was introduced when we switched to flat
metrics in Babel 2.

We now accept unfeasible updates when the source hasn't changed, just don't
update the route's timestamp.  This makes it more likely that we have
up-to-date information while still avoiding loops.

Triggered updates are no longer sent when a route gets better.

This slows down convergence a little, but reduces the amount of traffic.

Rely on periodic updates instead.

This is no longer needed now that we use a constant kernel priority.

This avoids searching for the source again.