This PR adds another probing routine to monitor the connection
for Raft message transports. Previously, we only monitored
snapshot transports.
In our production cluster, we found one TCP connection had >8-sec
latencies to a remote peer, but "etcd_network_peer_round_trip_time_seconds"
metrics shows <1-sec latency distribution, which means etcd server
was not sampling enough while such latency spikes happen
outside of snapshot pipeline connection.
Signed-off-by: Gyuho Lee <leegyuho@amazon.com>
