This is brittle, and prevents me from making route selection smarter.

This is the functionality described in Appendix C of RFC 6126.  Its
main purpose is to avoid keeping a full source table, which makes it
possible to implement a subset of Babel in just a few hundred lines of
code.  However, the code for maintaining the source table is already
there, and a parasitic implementation can be simulated using filtering
-- so it makes little sense to keep the functionality.

Thanks to Denis Ovsienko.

FreeBSD system headers have their own "struct route", which made it
impossible to compile babeld. Switching babeld to "struct babel_route".

We now limit resends to just 3 copies, both for requests and updates,
since there are enough mechanisms in the protocol to handle packet
loss.

They are no longer used except for configuring interfaces with no
explicit configuration.

This feature was little used, and complicated the code somewhat.

Nodes send wildcard requests at boot, which allows them to acquire
a full routing table in a timely manner.  Unfortunately, this causes
an update storm when a large network boots.

We used to allocate enough buffer space to send one full-size frame.
Unfortunately, in large networks this causes our update coalescing
strategy to fail, just where it is the most useful.  Allocate enough
buffers to fit a full update.

Yeah, it's a hack.

Conflicts:
	message.c

Reported by Thomas McLure.

This avoids the quadratic behaviour of expire_sources.

Reported by Thomas McLure.

Only apply the normal amount of hysteresis.

Reported by Dave Taht.

This changes the route table to be a sorted table of linked lists of
routes to a given prefix, which makes most route operations behave in
O(log n).  Insertion and flushing of a prefix is still O(n), but these
are fairly rare operations.

A nice side-effect is that the route table is now private to route.c,
which should make it easy to switch to a different data structure in
the future.