server: withdraw stale paths on RR cluster loops

[rastislavs]

This fixes an issue with stale paths when two route reflectors originate the same prefix and reflect routes to one another.

After they then both withdraw the local path, they can end up in a state with stale reflected routes on both RRs, even when none of them is advertising it anymore. See the attached reproducer: rr-self-reflect-repro.sh

RR-A originates 10.62.0.0/24
RR-B originates 10.62.0.0/24 before RR-A withdraws
--- RR-A global RIB
*> 10.62.0.0/24         0.0.0.0                                   00:00:04   [{Origin: ?}]
*  10.62.0.0/24         127.0.0.1                                 00:00:02   [{Origin: ?} {LocalPref: 100} {Originator: 10.254.0.9} {ClusterList: [255.0.0.2]}]
--- RR-B global RIB
*> 10.62.0.0/24         0.0.0.0                                   00:00:02   [{Origin: ?}]
*  10.62.0.0/24         127.0.0.1                                 00:00:04   [{Origin: ?} {LocalPref: 100} {Originator: 10.254.0.8} {ClusterList: [255.0.0.2]}]
RR-A withdraws its local path
--- RR-A global RIB
*> 10.62.0.0/24         127.0.0.1                                 00:00:04   [{Origin: ?} {LocalPref: 100} {Originator: 10.254.0.9} {ClusterList: [255.0.0.2]}]
--- RR-B global RIB
*> 10.62.0.0/24         0.0.0.0                                   00:00:04   [{Origin: ?}]
*  10.62.0.0/24         127.0.0.1                                 00:00:06   [{Origin: ?} {LocalPref: 100} {Originator: 10.254.0.8} {ClusterList: [255.0.0.2]}]
RR-B withdraws its local path
--- RR-A global RIB
*> 10.62.0.0/24         127.0.0.1                                 00:00:06   [{Origin: ?} {LocalPref: 100} {Originator: 10.254.0.9} {ClusterList: [255.0.0.2]}]
--- RR-B global RIB
*> 10.62.0.0/24         127.0.0.1                                 00:00:08   [{Origin: ?} {LocalPref: 100} {Originator: 10.254.0.8} {ClusterList: [255.0.0.2]}]
reproduced: both RRs retain reciprocal reflected paths after both local withdrawals