If the peer manager fails to find an available node during a call to selectNode(), it will latch a flag that can be used to trigger requesting more avalanche node addresses. The latch is atomic and cleared on read so we can loop over it in a network thread (out of scope for this diff).

Ref T1696.