bgpd: Modify the BGP to zebra route announcement to support multipath
routes. Use a growable buffer (bgp_nexthop_buf) to collect nexthops
that are included in the announcement. Use the BGP_INFO_MULTIPATH_CHG
flag to trigger zebra announcement so zebra will be updated if the
multipath set changes. Display all multipath nexthops in
'debug bgp zebra' output.

* bgpd/bgp_main.c
  * bgp_exit(): Free bgp_nexthop_buf when exiting
* bgpd/bgp_route.c
  * bgp_process_rsclient(): Clear BGP_INFO_MULTIPATH_CHG after processing
  * bgp_process_main(): Check BGP_INFO_MULTIPATH_CHG to trigger zebra
    announcement and clear aftr processing
* bgpd/bgp_zebra.c
  * bgp_nexthop_buf: Growable buffer used to collect nexthops for zebra
    announcement
  * bgp_zebra_announce(): Grow bgp_nexthop_buf if needed. Include
    multipath count in zebra announcement and add all nexthops to
    bgp_nexthop_buf. Pass bgp_nexthop_buf data to zebra announcement.
    Added nexthops to debug output.
  * bgp_zebra_init(): Initialize bgp_nexthop_buf at startup
* bgpd/bgp_zebra.h
  * BGP_NEXTHOP_BUF_SIZE: Default initial bgp_nexthop_buf size has room
    for 8 nexthops
diff --git a/bgpd/bgp_zebra.h b/bgpd/bgp_zebra.h
index 5172cb8..461255e 100644
--- a/bgpd/bgp_zebra.h
+++ b/bgpd/bgp_zebra.h
@@ -21,6 +21,10 @@
 #ifndef _QUAGGA_BGP_ZEBRA_H
 #define _QUAGGA_BGP_ZEBRA_H
 
+#define BGP_NEXTHOP_BUF_SIZE (8 * sizeof (struct in_addr *))
+
+extern struct stream *bgp_nexthop_buf;
+
 extern void bgp_zebra_init (void);
 extern int bgp_if_update_all (void);
 extern int bgp_config_write_maxpaths (struct vty *, struct bgp *, afi_t,