router: replica's prio is decreased on every replication flack

[Serpentian]

Currently router's failover checks replication state:

vshard/vshard/router/init.lua

Lines 1233 to 1240 in 85f9b66

	local status = upstream.status
	if not status or status == 'stopped' or status == 'disconnected' then
	-- All other states mean either that everything is ok ('follow')
	-- or that replica is connecting. In all these cases replica
	-- is considered healthy.
	local msg = string.format('Upstream to master has status "%s"', status)
	return consts.STATUS.RED, msg
	end

It doesn't pay attention to idle and lag if upstream is broken. This is not correct, since replica's priority will be changed on every disconnect and we should decrease replica's prio only after some timeout is passed. For that it's proposed to introduce failover_replica_lag_limit, which will check either idle or lag, depending on upstream.status. Initial proposal:

follow, sync => lag
all others => idle

[Serpentian]

Needed for #505